2022-07-14 13:18:57,860 fcos_core INFO: Using 1 GPUs 2022-07-14 13:18:57,860 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 13:18:57,860 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 13:19:01,735 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (6.2.1) 2022-07-14 13:19:01,735 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 13:19:01,736 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 2 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 13:19:01,737 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 4 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 2 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 13:19:07,650 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 13:19:07,675 fcos_core.trainer INFO: node_dis initialized 2022-07-14 13:19:07,676 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 13:19:07,678 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 13:19:08,023 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 13:19:08,024 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 13:19:08,090 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 13:19:08,090 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 13:19:08,090 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 13:19:08,090 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 13:19:08,090 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 13:19:08,091 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 13:19:08,091 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 13:19:08,091 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 13:19:08,091 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 13:19:08,091 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 13:19:08,091 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 13:19:08,091 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 13:19:08,091 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 13:19:08,091 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 13:19:08,092 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 13:19:08,092 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 13:19:08,092 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 13:19:08,092 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 13:19:08,092 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 13:19:08,092 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 13:19:08,092 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 13:19:08,092 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 13:19:08,092 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 13:19:08,092 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 13:19:08,093 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 13:19:08,093 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 13:19:09,360 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 13:19:13,373 fcos_core.trainer INFO: Start training 2022-07-14 13:21:57,919 fcos_core INFO: Using 1 GPUs 2022-07-14 13:21:57,919 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 13:21:57,919 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 13:22:01,926 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.2.1 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.2.1 pypi_0 pypi Pillow (6.2.1) 2022-07-14 13:22:01,927 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 13:22:01,927 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 2 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 13:22:01,930 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 4 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 2 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 13:22:07,025 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 13:22:07,049 fcos_core.trainer INFO: node_dis initialized 2022-07-14 13:22:07,051 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 13:22:07,052 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 13:22:07,385 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 13:22:07,385 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 13:22:07,445 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 13:22:07,445 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 13:22:07,445 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 13:22:07,446 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 13:22:07,446 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 13:22:07,446 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 13:22:07,446 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 13:22:07,446 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 13:22:07,446 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 13:22:07,446 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 13:22:07,446 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 13:22:07,446 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 13:22:07,446 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 13:22:07,447 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 13:22:07,447 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 13:22:07,447 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 13:22:07,447 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 13:22:07,447 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 13:22:07,448 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 13:22:07,448 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 13:22:07,448 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 13:22:07,448 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 13:22:07,448 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 13:22:07,448 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 13:22:07,448 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 13:22:07,449 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 13:22:08,856 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 13:22:12,449 fcos_core.trainer INFO: Start training 2022-07-14 13:28:50,198 fcos_core INFO: Using 1 GPUs 2022-07-14 13:28:50,198 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 13:28:50,198 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 13:28:54,147 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.2.1 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.2.1 pypi_0 pypi Pillow (6.2.1) 2022-07-14 13:28:54,147 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 13:28:54,148 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 13:28:54,149 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 4 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 13:28:59,636 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 13:28:59,662 fcos_core.trainer INFO: node_dis initialized 2022-07-14 13:28:59,664 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 13:28:59,665 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 13:29:00,055 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 13:29:00,055 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 13:29:00,120 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 13:29:00,121 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 13:29:00,121 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 13:29:00,122 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 13:29:00,122 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 13:29:00,122 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 13:29:00,122 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 13:29:00,122 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 13:29:00,122 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 13:29:00,122 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 13:29:00,122 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 13:29:00,122 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 13:29:00,122 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 13:29:00,122 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 13:29:00,122 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 13:29:00,123 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 13:29:00,123 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 13:29:00,123 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 13:29:00,123 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 13:29:00,123 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 13:29:00,123 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 13:29:00,123 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 13:29:00,123 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 13:29:00,123 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 13:29:00,123 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 13:29:00,123 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 13:29:01,389 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 13:29:05,494 fcos_core.trainer INFO: Start training 2022-07-14 13:39:49,311 fcos_core INFO: Using 1 GPUs 2022-07-14 13:39:49,311 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 13:39:49,312 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 13:39:53,343 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.2.1 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.2.1 pypi_0 pypi Pillow (6.2.1) 2022-07-14 13:39:53,343 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 13:39:53,343 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 13:39:53,345 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 2 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 13:39:58,480 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 13:39:58,508 fcos_core.trainer INFO: node_dis initialized 2022-07-14 13:39:58,509 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 13:39:58,510 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 13:39:58,906 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 13:39:58,907 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 13:39:58,965 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 13:39:58,966 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 13:39:58,966 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 13:39:58,966 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 13:39:58,966 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 13:39:58,966 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 13:39:58,966 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 13:39:58,966 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 13:39:58,966 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 13:39:58,966 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 13:39:58,966 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 13:39:58,966 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 13:39:58,966 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 13:39:58,967 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 13:39:58,967 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 13:39:58,967 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 13:39:58,967 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 13:39:58,967 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 13:39:58,967 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 13:39:58,967 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 13:39:58,967 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 13:39:58,967 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 13:39:58,967 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 13:39:58,967 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 13:39:58,968 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 13:39:58,968 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 13:40:00,309 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 13:40:03,802 fcos_core.trainer INFO: Start training 2022-07-14 13:40:38,561 fcos_core INFO: Using 1 GPUs 2022-07-14 13:40:38,561 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 13:40:38,562 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 13:40:42,557 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.2.1 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.2.1 pypi_0 pypi Pillow (6.2.1) 2022-07-14 13:40:42,558 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 13:40:42,558 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 13:40:42,562 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 13:40:47,920 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 13:40:47,951 fcos_core.trainer INFO: node_dis initialized 2022-07-14 13:40:47,953 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 13:40:47,954 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 13:40:48,359 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 13:40:48,359 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 13:40:48,421 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 13:40:48,422 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 13:40:48,422 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 13:40:48,422 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 13:40:48,422 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 13:40:48,423 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 13:40:48,423 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 13:40:48,423 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 13:40:48,423 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 13:40:48,423 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 13:40:48,423 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 13:40:48,423 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 13:40:48,423 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 13:40:48,423 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 13:40:48,423 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 13:40:48,424 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 13:40:48,424 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 13:40:48,424 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 13:40:48,424 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 13:40:48,424 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 13:40:48,424 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 13:40:48,424 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 13:40:48,424 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 13:40:48,424 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 13:40:48,424 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 13:40:48,425 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 13:40:49,776 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 13:40:53,399 fcos_core.trainer INFO: Start training 2022-07-14 13:53:14,993 fcos_core INFO: Using 1 GPUs 2022-07-14 13:53:14,993 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 13:53:15,002 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 13:53:19,137 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (6.2.1) 2022-07-14 13:53:19,137 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 13:53:19,137 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 13:53:19,140 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 13:53:24,241 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 13:53:24,266 fcos_core.trainer INFO: node_dis initialized 2022-07-14 13:53:24,267 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 13:53:24,269 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 13:53:24,609 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 13:53:24,609 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 13:53:24,664 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 13:53:24,664 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 13:53:24,664 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 13:53:24,664 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 13:53:24,664 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 13:53:24,664 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 13:53:24,665 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 13:53:24,665 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 13:53:24,665 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 13:53:24,665 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 13:53:24,665 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 13:53:24,665 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 13:53:24,665 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 13:53:24,665 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 13:53:24,665 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 13:53:24,665 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 13:53:24,665 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 13:53:24,665 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 13:53:24,666 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 13:53:24,666 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 13:53:24,666 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 13:53:24,666 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 13:53:24,666 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 13:53:24,666 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 13:53:24,666 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 13:53:24,666 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 13:53:25,933 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 13:53:29,830 fcos_core.trainer INFO: Start training 2022-07-14 13:57:47,335 fcos_core INFO: Using 1 GPUs 2022-07-14 13:57:47,336 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 13:57:47,336 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 13:57:51,385 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (6.2.1) 2022-07-14 13:57:51,386 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 13:57:51,386 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 13:57:51,389 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 13:57:56,780 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 13:57:56,802 fcos_core.trainer INFO: node_dis initialized 2022-07-14 13:57:56,804 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 13:57:56,805 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 13:57:57,240 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 13:57:57,241 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 13:57:57,310 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 13:57:57,310 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 13:57:57,310 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 13:57:57,310 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 13:57:57,311 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 13:57:57,311 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 13:57:57,311 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 13:57:57,311 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 13:57:57,311 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 13:57:57,311 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 13:57:57,311 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 13:57:57,311 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 13:57:57,311 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 13:57:57,311 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 13:57:57,311 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 13:57:57,311 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 13:57:57,311 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 13:57:57,312 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 13:57:57,312 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 13:57:57,312 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 13:57:57,312 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 13:57:57,312 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 13:57:57,312 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 13:57:57,312 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 13:57:57,312 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 13:57:57,312 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 13:57:58,702 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 13:58:02,686 fcos_core.trainer INFO: Start training 2022-07-14 14:09:26,670 fcos_core INFO: Using 1 GPUs 2022-07-14 14:09:26,671 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 14:09:26,671 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 14:09:30,485 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (6.2.1) 2022-07-14 14:09:30,485 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 14:09:30,486 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 14:09:30,487 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 14:09:35,846 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 14:09:35,879 fcos_core.trainer INFO: node_dis initialized 2022-07-14 14:09:35,881 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 14:09:35,883 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 14:09:36,262 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 14:09:36,262 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 14:09:36,315 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 14:09:36,316 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 14:09:36,316 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 14:09:36,316 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 14:09:36,316 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 14:09:36,316 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 14:09:36,317 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 14:09:36,317 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 14:09:36,317 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 14:09:36,317 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 14:09:36,317 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 14:09:36,317 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 14:09:36,318 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 14:09:36,318 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 14:09:36,318 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 14:09:36,318 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 14:09:36,318 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 14:09:36,318 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 14:09:36,318 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 14:09:36,318 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 14:09:36,318 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 14:09:36,318 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 14:09:36,318 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 14:09:36,319 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 14:09:36,319 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 14:09:36,319 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 14:09:37,641 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 14:09:41,093 fcos_core.trainer INFO: Start training 2022-07-14 14:11:57,009 fcos_core INFO: Using 1 GPUs 2022-07-14 14:11:57,009 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 14:11:57,010 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 14:12:01,178 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (6.2.1) 2022-07-14 14:12:01,179 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 14:12:01,179 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 14:12:01,181 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 14:12:06,530 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 14:12:06,557 fcos_core.trainer INFO: node_dis initialized 2022-07-14 14:12:06,559 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 14:12:06,560 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 14:12:07,000 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 14:12:07,001 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 14:12:07,072 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 14:12:07,073 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 14:12:07,073 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 14:12:07,073 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 14:12:07,073 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 14:12:07,073 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 14:12:07,073 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 14:12:07,073 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 14:12:07,073 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 14:12:07,074 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 14:12:07,074 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 14:12:07,074 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 14:12:07,074 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 14:12:07,074 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 14:12:07,074 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 14:12:07,074 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 14:12:07,074 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 14:12:07,074 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 14:12:07,074 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 14:12:07,075 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 14:12:07,075 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 14:12:07,075 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 14:12:07,075 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 14:12:07,075 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 14:12:07,075 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 14:12:07,075 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 14:12:08,312 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 14:12:12,358 fcos_core.trainer INFO: Start training 2022-07-14 14:20:45,897 fcos_core INFO: Using 1 GPUs 2022-07-14 14:20:45,897 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 14:20:45,898 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 14:20:49,942 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (6.2.1) 2022-07-14 14:20:49,943 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 14:20:49,943 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 14:20:49,947 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 14:20:55,271 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 14:20:55,304 fcos_core.trainer INFO: node_dis initialized 2022-07-14 14:20:55,306 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 14:20:55,308 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 14:20:55,770 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 14:20:55,770 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 14:20:55,838 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 14:20:55,839 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 14:20:55,839 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 14:20:55,839 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 14:20:55,839 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 14:20:55,839 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 14:20:55,839 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 14:20:55,839 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 14:20:55,839 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 14:20:55,840 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 14:20:55,840 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 14:20:55,840 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 14:20:55,840 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 14:20:55,840 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 14:20:55,840 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 14:20:55,840 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 14:20:55,840 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 14:20:55,840 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 14:20:55,840 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 14:20:55,840 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 14:20:55,841 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 14:20:55,841 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 14:20:55,841 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 14:20:55,841 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 14:20:55,841 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 14:20:55,841 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 14:20:57,155 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 14:21:01,224 fcos_core.trainer INFO: Start training 2022-07-14 14:25:39,588 fcos_core INFO: Using 1 GPUs 2022-07-14 14:25:39,588 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 14:25:39,588 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 14:25:43,484 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (6.2.1) 2022-07-14 14:25:43,484 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 14:25:43,485 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 14:25:43,486 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 14:25:48,912 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 14:25:48,946 fcos_core.trainer INFO: node_dis initialized 2022-07-14 14:25:48,948 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 14:25:48,949 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 14:25:49,306 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 14:25:49,307 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 14:25:49,367 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 14:25:49,367 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 14:25:49,367 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 14:25:49,367 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 14:25:49,367 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 14:25:49,367 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 14:25:49,367 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 14:25:49,367 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 14:25:49,368 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 14:25:49,368 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 14:25:49,368 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 14:25:49,368 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 14:25:49,368 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 14:25:49,368 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 14:25:49,368 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 14:25:49,368 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 14:25:49,368 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 14:25:49,368 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 14:25:49,368 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 14:25:49,368 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 14:25:49,369 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 14:25:49,369 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 14:25:49,369 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 14:25:49,369 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 14:25:49,369 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 14:25:49,369 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 14:25:50,702 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 14:25:54,416 fcos_core.trainer INFO: Start training 2022-07-14 14:27:34,032 fcos_core INFO: Using 1 GPUs 2022-07-14 14:27:34,033 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 14:27:34,033 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 14:27:38,304 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 14:27:38,305 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 14:27:38,305 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 14:27:38,307 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 14:39:16,632 fcos_core INFO: Using 1 GPUs 2022-07-14 14:39:16,633 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 14:39:16,633 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 14:39:20,784 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 14:39:20,785 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 14:39:20,785 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 14:39:20,786 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda:4 FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 14:39:26,233 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 14:39:26,258 fcos_core.trainer INFO: node_dis initialized 2022-07-14 14:39:26,259 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 14:39:26,260 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 14:39:26,647 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 14:39:26,647 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 14:39:26,711 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 14:39:26,711 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 14:39:26,711 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 14:39:26,711 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 14:39:26,711 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 14:39:26,711 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 14:39:26,712 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 14:39:26,712 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 14:39:26,712 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 14:39:26,712 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 14:39:26,712 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 14:39:26,712 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 14:39:26,712 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 14:39:26,713 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 14:39:26,713 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 14:39:26,713 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 14:39:26,713 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 14:39:26,713 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 14:39:26,713 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 14:39:26,713 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 14:39:26,714 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 14:39:26,714 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 14:39:26,714 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 14:39:26,714 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 14:39:26,714 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 14:39:26,714 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 14:39:28,031 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 14:39:32,088 fcos_core.trainer INFO: Start training 2022-07-14 14:48:09,802 fcos_core INFO: Using 1 GPUs 2022-07-14 14:48:09,802 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 14:48:09,802 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 14:48:13,878 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 14:48:13,878 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 14:48:13,879 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 14:48:13,880 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda:4 FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 14:48:19,499 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 14:48:19,522 fcos_core.trainer INFO: node_dis initialized 2022-07-14 14:48:19,524 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 14:48:19,525 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 14:48:19,928 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 14:48:19,929 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 14:48:19,995 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 14:48:19,996 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 14:48:19,996 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 14:48:19,996 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 14:48:19,996 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 14:48:19,997 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 14:48:19,997 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 14:48:19,997 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 14:48:19,997 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 14:48:19,997 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 14:48:19,997 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 14:48:19,997 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 14:48:19,997 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 14:48:19,997 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 14:48:19,997 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 14:48:19,997 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 14:48:19,997 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 14:48:19,997 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 14:48:19,998 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 14:48:19,998 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 14:48:19,998 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 14:48:19,998 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 14:48:19,998 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 14:48:19,998 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 14:48:19,998 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 14:48:19,998 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 14:48:21,319 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 14:48:26,352 fcos_core.trainer INFO: Start training 2022-07-14 14:51:37,969 fcos_core INFO: Using 1 GPUs 2022-07-14 14:51:37,970 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 14:51:37,970 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 14:51:42,117 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 14:51:42,118 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 14:51:42,118 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 14:51:42,121 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda:4 FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 14:51:47,766 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 14:51:47,790 fcos_core.trainer INFO: node_dis initialized 2022-07-14 14:51:47,792 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 14:51:47,793 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 14:51:48,205 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 14:51:48,206 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 14:51:48,273 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 14:51:48,273 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 14:51:48,273 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 14:51:48,274 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 14:51:48,274 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 14:51:48,274 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 14:51:48,274 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 14:51:48,274 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 14:51:48,274 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 14:51:48,274 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 14:51:48,274 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 14:51:48,274 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 14:51:48,274 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 14:51:48,274 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 14:51:48,274 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 14:51:48,274 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 14:51:48,275 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 14:51:48,275 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 14:51:48,275 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 14:51:48,275 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 14:51:48,275 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 14:51:48,275 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 14:51:48,275 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 14:51:48,275 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 14:51:48,275 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 14:51:48,275 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 14:51:49,629 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 14:51:53,626 fcos_core.trainer INFO: Start training 2022-07-14 14:54:21,280 fcos_core INFO: Using 1 GPUs 2022-07-14 14:54:21,280 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 14:54:21,281 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 14:54:25,443 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 14:54:25,443 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 14:54:25,444 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 14:54:25,446 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda:4 FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 14:54:31,890 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 14:54:31,915 fcos_core.trainer INFO: node_dis initialized 2022-07-14 14:54:31,916 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 14:54:31,918 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 14:54:32,254 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 14:54:32,255 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 14:54:32,324 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 14:54:32,324 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 14:54:32,325 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 14:54:32,325 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 14:54:32,325 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 14:54:32,325 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 14:54:32,325 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 14:54:32,325 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 14:54:32,325 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 14:54:32,326 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 14:54:32,326 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 14:54:32,326 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 14:54:32,326 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 14:54:32,326 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 14:54:32,326 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 14:54:32,326 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 14:54:32,326 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 14:54:32,326 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 14:54:32,326 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 14:54:32,326 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 14:54:32,327 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 14:54:32,327 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 14:54:32,327 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 14:54:32,327 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 14:54:32,327 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 14:54:32,327 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 14:54:33,718 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 14:54:38,595 fcos_core.trainer INFO: Start training 2022-07-14 15:02:11,766 fcos_core INFO: Using 1 GPUs 2022-07-14 15:02:11,767 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 15:02:11,767 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 15:02:15,722 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 15:02:15,724 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 15:02:15,724 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 15:02:15,728 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda:4 FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 15:02:21,439 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 15:02:21,463 fcos_core.trainer INFO: node_dis initialized 2022-07-14 15:02:21,465 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 15:02:21,466 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 15:02:21,842 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 15:02:21,843 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 15:02:21,906 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 15:02:21,907 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 15:02:21,907 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 15:02:21,907 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 15:02:21,907 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 15:02:21,907 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 15:02:21,907 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 15:02:21,907 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 15:02:21,907 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 15:02:21,907 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 15:02:21,907 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 15:02:21,908 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 15:02:21,908 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 15:02:21,908 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 15:02:21,908 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 15:02:21,908 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 15:02:21,908 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 15:02:21,908 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 15:02:21,908 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 15:02:21,908 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 15:02:21,908 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 15:02:21,908 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 15:02:21,908 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 15:02:21,909 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 15:02:21,909 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 15:02:21,909 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 15:02:23,321 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 15:02:28,406 fcos_core.trainer INFO: Start training 2022-07-14 15:03:59,807 fcos_core INFO: Using 1 GPUs 2022-07-14 15:03:59,807 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 15:03:59,807 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 15:04:03,911 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 15:04:03,912 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 15:04:03,913 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 15:04:03,916 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda:4 FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 15:04:10,336 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 15:04:10,360 fcos_core.trainer INFO: node_dis initialized 2022-07-14 15:04:10,361 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 15:04:10,362 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 15:04:10,866 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 15:04:10,867 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 15:04:10,932 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 15:04:10,932 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 15:04:10,932 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 15:04:10,932 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 15:04:10,932 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 15:04:10,932 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 15:04:10,933 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 15:04:10,933 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 15:04:10,933 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 15:04:10,933 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 15:04:10,933 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 15:04:10,933 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 15:04:10,933 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 15:04:10,934 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 15:04:10,934 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 15:04:10,934 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 15:04:10,934 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 15:04:10,934 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 15:04:10,934 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 15:04:10,934 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 15:04:10,934 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 15:04:10,934 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 15:04:10,934 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 15:04:10,935 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 15:04:10,935 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 15:04:10,935 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 15:04:12,220 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 15:04:16,805 fcos_core.trainer INFO: Start training 2022-07-14 15:05:59,947 fcos_core INFO: Using 1 GPUs 2022-07-14 15:05:59,947 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 15:05:59,947 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 15:06:04,041 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 15:06:04,042 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 15:06:04,042 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 15:06:04,044 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 15:06:09,410 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 15:06:09,434 fcos_core.trainer INFO: node_dis initialized 2022-07-14 15:06:09,435 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 15:06:09,436 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 15:06:09,778 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 15:06:09,778 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 15:06:09,832 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 15:06:09,833 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 15:06:09,833 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 15:06:09,833 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 15:06:09,833 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 15:06:09,833 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 15:06:09,833 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 15:06:09,833 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 15:06:09,833 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 15:06:09,834 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 15:06:09,834 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 15:06:09,834 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 15:06:09,834 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 15:06:09,834 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 15:06:09,834 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 15:06:09,834 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 15:06:09,834 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 15:06:09,834 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 15:06:09,834 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 15:06:09,835 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 15:06:09,835 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 15:06:09,835 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 15:06:09,835 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 15:06:09,835 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 15:06:09,835 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 15:06:09,835 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 15:06:11,060 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 15:06:15,189 fcos_core.trainer INFO: Start training 2022-07-14 15:20:51,651 fcos_core INFO: Using 1 GPUs 2022-07-14 15:20:51,651 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 15:20:51,652 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 15:20:55,773 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 15:20:55,774 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 15:20:55,774 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 15:20:55,778 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda:4 FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 15:21:02,499 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 15:21:02,534 fcos_core.trainer INFO: node_dis initialized 2022-07-14 15:21:02,536 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 15:21:02,538 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 15:21:02,946 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 15:21:02,946 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 15:21:03,019 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 15:21:03,020 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 15:21:03,020 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 15:21:03,020 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 15:21:03,020 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 15:21:03,020 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 15:21:03,020 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 15:21:03,020 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 15:21:03,020 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 15:21:03,020 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 15:21:03,020 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 15:21:03,020 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 15:21:03,020 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 15:21:03,021 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 15:21:03,021 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 15:21:03,021 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 15:21:03,021 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 15:21:03,021 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 15:21:03,021 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 15:21:03,021 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 15:21:03,021 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 15:21:03,021 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 15:21:03,021 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 15:21:03,021 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 15:21:03,022 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 15:21:03,022 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 15:21:04,474 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 15:21:10,197 fcos_core.trainer INFO: Start training 2022-07-14 15:23:30,929 fcos_core INFO: Using 1 GPUs 2022-07-14 15:23:30,929 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=4, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 15:23:30,929 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 15:23:35,158 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 15:23:35,159 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 15:23:35,159 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 15:23:35,161 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda:4 FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 15:23:40,977 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 15:23:41,003 fcos_core.trainer INFO: node_dis initialized 2022-07-14 15:23:41,005 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 15:23:41,006 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 15:23:41,407 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 15:23:41,408 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 15:23:41,479 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 15:23:41,479 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 15:23:41,479 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 15:23:41,480 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 15:23:41,480 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 15:23:41,480 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 15:23:41,480 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 15:23:41,480 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 15:23:41,480 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 15:23:41,480 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 15:23:41,480 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 15:23:41,480 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 15:23:41,480 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 15:23:41,481 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 15:23:41,481 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 15:23:41,481 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 15:23:41,482 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 15:23:41,482 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 15:23:41,482 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 15:23:41,482 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 15:23:41,482 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 15:23:41,482 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 15:23:41,482 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 15:23:41,482 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 15:23:41,483 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 15:23:41,483 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 15:23:42,690 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 15:23:47,353 fcos_core.trainer INFO: Start training 2022-07-14 15:31:54,933 fcos_core INFO: Using 1 GPUs 2022-07-14 15:31:54,934 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=4, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 15:31:54,934 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 15:31:58,781 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 15:31:58,781 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 15:31:58,781 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 15:31:58,784 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda:4 FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 15:32:04,800 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 15:32:04,825 fcos_core.trainer INFO: node_dis initialized 2022-07-14 15:32:04,826 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 15:32:04,828 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 15:32:05,160 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 15:32:05,161 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 15:32:05,223 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 15:32:05,223 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 15:32:05,223 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 15:32:05,223 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 15:32:05,223 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 15:32:05,223 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 15:32:05,224 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 15:32:05,224 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 15:32:05,224 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 15:32:05,224 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 15:32:05,224 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 15:32:05,224 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 15:32:05,224 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 15:32:05,224 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 15:32:05,224 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 15:32:05,224 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 15:32:05,224 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 15:32:05,225 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 15:32:05,225 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 15:32:05,225 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 15:32:05,225 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 15:32:05,225 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 15:32:05,225 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 15:32:05,225 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 15:32:05,225 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 15:32:05,225 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 15:32:06,573 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 15:32:11,029 fcos_core.trainer INFO: Start training 2022-07-14 15:52:53,923 fcos_core INFO: Using 1 GPUs 2022-07-14 15:52:53,924 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 15:52:53,924 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 15:52:58,183 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 15:52:58,186 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 15:52:58,186 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 1 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 15:52:58,189 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 1 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 1 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 15:53:03,961 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 15:53:03,993 fcos_core.trainer INFO: node_dis initialized 2022-07-14 15:53:03,995 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 15:53:03,997 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 15:53:04,433 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 15:53:04,434 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 15:53:04,494 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 15:53:04,495 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 15:53:04,495 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 15:53:04,495 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 15:53:04,495 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 15:53:04,496 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 15:53:04,496 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 15:53:04,496 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 15:53:04,496 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 15:53:04,496 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 15:53:04,497 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 15:53:04,497 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 15:53:04,497 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 15:53:04,497 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 15:53:04,497 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 15:53:04,497 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 15:53:04,498 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 15:53:04,498 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 15:53:04,498 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 15:53:04,498 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 15:53:04,498 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 15:53:04,499 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 15:53:04,499 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 15:53:04,499 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 15:53:04,499 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 15:53:04,499 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 15:53:05,739 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 15:53:10,454 fcos_core.trainer INFO: Start training 2022-07-14 15:53:35,013 fcos_core.trainer INFO: eta: 2 days, 20:12:06 iter: 20 loss_ds: 5.5346 (5.5706) node_loss: 0.6667 (0.6669) loss_cls: 0.5915 (0.7554) loss_reg: 2.5164 (2.7899) loss_centerness: 0.6656 (0.6742) loss_adv_P7: 0.1397 (0.1400) loss_adv_P6: 0.1400 (0.1402) loss_adv_P5: 0.1400 (0.1406) loss_adv_P4: 0.1371 (0.1375) loss_adv_P3: 0.1380 (0.1380) time: 1.0710 (1.2278) data: 0.0090 (0.0371) dis_loss: 0.0707 (0.0711) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 1989 2022-07-14 15:54:03,557 fcos_core.trainer INFO: eta: 3 days, 1:44:04 iter: 40 loss_ds: 6.1282 (5.9188) node_loss: 0.6211 (0.6433) loss_cls: 2.3082 (1.6360) loss_reg: 1.7141 (2.2951) loss_centerness: 0.6647 (0.6670) loss_adv_P7: 0.1378 (0.1391) loss_adv_P6: 0.1382 (0.1390) loss_adv_P5: 0.1369 (0.1386) loss_adv_P4: 0.1270 (0.1322) loss_adv_P3: 0.1271 (0.1320) time: 1.4409 (1.3275) data: 0.0133 (0.0252) dis_loss: 0.0723 (0.0716) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 1989 2022-07-14 15:54:27,908 fcos_core.trainer INFO: eta: 2 days, 23:41:29 iter: 60 loss_ds: 4.6701 (5.5201) node_loss: 0.5557 (0.6137) loss_cls: 1.2167 (1.4840) loss_reg: 1.7018 (2.1184) loss_centerness: 0.6628 (0.6656) loss_adv_P7: 0.1254 (0.1340) loss_adv_P6: 0.1255 (0.1333) loss_adv_P5: 0.1224 (0.1326) loss_adv_P4: 0.0934 (0.1196) loss_adv_P3: 0.0985 (0.1208) time: 1.2732 (1.2908) data: 0.0121 (0.0210) dis_loss: 0.0723 (0.0716) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 1989 2022-07-14 15:56:23,680 fcos_core INFO: Using 1 GPUs 2022-07-14 15:56:23,680 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 15:56:23,680 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 15:56:27,720 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 15:56:27,721 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 15:56:27,721 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'none' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 200000 # 4 for source and 4 for target IMS_PER_BATCH: 2 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 15:56:27,724 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 4 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: none MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 50 NUM_NODES_PER_LVL_TG: 50 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 2 INITIAL_AP50: 35 MAX_ITER: 200000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 15:56:34,475 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 15:56:34,506 fcos_core.trainer INFO: node_dis initialized 2022-07-14 15:56:34,507 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 15:56:34,508 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 15:56:34,849 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 15:56:34,850 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 15:56:34,925 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 15:56:34,925 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 15:56:34,925 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 15:56:34,925 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 15:56:34,925 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 15:56:34,925 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 15:56:34,925 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 15:56:34,926 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 15:56:34,926 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 15:56:34,926 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 15:56:34,926 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 15:56:34,926 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 15:56:34,926 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 15:56:34,926 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 15:56:34,926 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 15:56:34,926 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 15:56:34,926 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 15:56:34,926 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 15:56:34,927 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 15:56:34,927 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 15:56:34,927 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 15:56:34,927 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 15:56:34,927 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 15:56:34,927 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 15:56:34,927 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 15:56:34,927 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 15:56:36,204 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 15:56:40,644 fcos_core.trainer INFO: Start training 2022-07-14 15:57:23,108 fcos_core.trainer INFO: eta: 4 days, 21:55:47 iter: 20 loss_ds: 6.7936 (6.8332) node_loss: 0.6641 (0.6662) loss_cls: 2.2057 (2.1675) loss_reg: 2.1425 (2.6181) loss_centerness: 0.6721 (0.6747) loss_adv_P7: 0.1395 (0.1397) loss_adv_P6: 0.1400 (0.1400) loss_adv_P5: 0.1401 (0.1405) loss_adv_P4: 0.1373 (0.1378) loss_adv_P3: 0.1380 (0.1382) time: 1.9026 (2.1230) data: 0.0101 (0.0848) dis_loss: 0.0700 (0.0707) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 3552 2022-07-14 15:58:02,822 fcos_core.trainer INFO: eta: 4 days, 18:06:26 iter: 40 loss_ds: 7.9626 (8.0694) node_loss: 0.6246 (0.6444) loss_cls: 3.9928 (3.8549) loss_reg: 1.7874 (2.2175) loss_centerness: 0.6499 (0.6645) loss_adv_P7: 0.1364 (0.1382) loss_adv_P6: 0.1361 (0.1380) loss_adv_P5: 0.1359 (0.1381) loss_adv_P4: 0.1309 (0.1344) loss_adv_P3: 0.1273 (0.1325) time: 1.9448 (2.0543) data: 0.0189 (0.0522) dis_loss: 0.0699 (0.0705) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 3552 2022-07-14 15:58:41,262 fcos_core.trainer INFO: eta: 4 days, 15:38:42 iter: 60 loss_ds: 5.4760 (7.9245) node_loss: 0.5998 (0.6297) loss_cls: 1.5723 (3.9330) loss_reg: 1.5699 (2.0254) loss_centerness: 0.6529 (0.6615) loss_adv_P7: 0.1352 (0.1374) loss_adv_P6: 0.1372 (0.1376) loss_adv_P5: 0.1348 (0.1369) loss_adv_P4: 0.1241 (0.1310) loss_adv_P3: 0.1174 (0.1272) time: 1.8670 (2.0102) data: 0.0212 (0.0417) dis_loss: 0.0699 (0.0705) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 3552 2022-07-14 16:00:38,407 fcos_core INFO: Using 1 GPUs 2022-07-14 16:00:38,407 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 16:00:38,408 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 16:00:42,402 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 16:00:42,404 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 16:00:42,404 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'o2o' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 100 NUM_NODES_PER_LVL_TG: 100 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 100000 # 4 for source and 4 for target IMS_PER_BATCH: 4 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 16:00:42,408 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 4 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: o2o MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 100 NUM_NODES_PER_LVL_TG: 100 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 4 INITIAL_AP50: 35 MAX_ITER: 100000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 16:00:48,554 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 16:00:48,577 fcos_core.trainer INFO: node_dis initialized 2022-07-14 16:00:48,578 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 16:00:48,580 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 16:00:48,891 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 16:00:48,892 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 16:00:48,942 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 16:00:48,943 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 16:00:48,943 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 16:00:48,943 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 16:00:48,944 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 16:00:48,944 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 16:00:48,944 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 16:00:48,944 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 16:00:48,944 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 16:00:48,945 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 16:00:48,945 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 16:00:48,945 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 16:00:48,945 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 16:00:48,945 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 16:00:48,945 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 16:00:48,945 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 16:00:48,946 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 16:00:48,946 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 16:00:48,946 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 16:00:48,946 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 16:00:48,946 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 16:00:48,946 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 16:00:48,946 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 16:00:48,946 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 16:00:48,946 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 16:00:48,946 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 16:00:50,291 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 16:00:55,020 fcos_core.trainer INFO: Start training 2022-07-14 16:06:00,721 fcos_core INFO: Using 1 GPUs 2022-07-14 16:06:00,722 fcos_core INFO: Namespace(config_file='configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml', distributed=False, local_rank=0, opts=[], skip_test=False, test_only=False, use_tensorboard=True) 2022-07-14 16:06:00,722 fcos_core INFO: Collecting env info (might take some time) 2022-07-14 16:06:04,728 fcos_core INFO: PyTorch version: 1.10.0 Is debug build: False CUDA used to build PyTorch: 11.3 ROCM used to build PyTorch: N/A OS: Ubuntu 18.04.6 LTS (x86_64) GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.10 Python version: 3.7.9 (default, Aug 31 2020, 12:42:55) [GCC 7.3.0] (64-bit runtime) Python platform: Linux-5.4.0-99-generic-x86_64-with-debian-buster-sid Is CUDA available: True CUDA runtime version: Could not collect GPU models and configuration: GPU 0: NVIDIA RTX A4000 GPU 1: NVIDIA RTX A4000 GPU 2: NVIDIA RTX A4000 GPU 3: NVIDIA RTX A4000 GPU 4: NVIDIA RTX A4000 GPU 5: NVIDIA RTX A4000 GPU 6: NVIDIA RTX A4000 GPU 7: NVIDIA RTX A4000 Nvidia driver version: 470.86 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Versions of relevant libraries: [pip3] numpy==1.21.6 [pip3] torch==1.10.0 [pip3] torchaudio==0.10.0 [pip3] torchvision==0.11.0 [conda] blas 1.0 mkl defaults [conda] cudatoolkit 11.3.1 h2bc3f7f_2 defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2021.4.0 h06a4308_640 defaults [conda] mkl-service 2.4.0 py37h402132d_0 conda-forge [conda] mkl_fft 1.3.1 py37h3e078e5_1 conda-forge [conda] mkl_random 1.2.2 py37h219a48f_0 conda-forge [conda] numpy 1.21.6 pypi_0 pypi [conda] numpy-base 1.21.5 py37ha15fc14_3 defaults [conda] pytorch 1.10.0 py3.7_cuda11.3_cudnn8.2.0_0 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] torchaudio 0.10.0 py37_cu113 pytorch [conda] torchvision 0.11.0 py37_cu113 pytorch Pillow (9.2.0) 2022-07-14 16:06:04,730 fcos_core INFO: Loaded configuration file configs/SIGMA/sigma_vgg16_sim10k_to_cityscapes.yaml 2022-07-14 16:06:04,730 fcos_core INFO: OUTPUT_DIR: './experiments/sigma/sim10k_to_cityscapes_vgg16/' MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth' # Initialed by imagenet # NOTE: In our cvpr version, we mistakely set FCOS.NMS_TH as 0.3, giving a 53.7 result. After setting it correctly, sigma gives 57.1 mAP.... # WEIGHT: './well_trained_models/sim10k_to_city_vgg16_53.73_mAP.pth' # Initialed by pretrained weight RPN_ONLY: True FCOS_ON: True DA_ON: True ATSS_ON: False MIDDLE_HEAD_CFG: 'GM_HEAD' MIDDLE_HEAD: CONDGRAPH_ON: True IN_NORM: 'LN' NUM_CONVS_IN: 2 GM: # matching cfg MATCHING_LOSS_CFG: 'MSE' MATCHING_CFG: 'o2o' WITH_SCORE_WEIGHT: False WITH_NODE_DIS: True # node sampling NUM_NODES_PER_LVL_SR: 100 NUM_NODES_PER_LVL_TG: 100 BG_RATIO: 2 # loss weight MATCHING_LOSS_WEIGHT: 1.0 NODE_LOSS_WEIGHT: 1.0 NODE_DIS_WEIGHT: 0.1 NODE_DIS_LAMBDA: 0.02 WITH_SEMANTIC_COMPLETION: True WITH_QUADRATIC_MATCHING: True WITH_CLUSTER_UPDATE: True WITH_CTR: False WITH_COMPLETE_GRAPH: True WITH_DOMAIN_INTERACTION: True BACKBONE: CONV_BODY: "VGG-16-FPN-RETINANET" RETINANET: USE_C5: False # FCOS uses P5 instead of C5 FCOS: NUM_CONVS_REG: 4 NUM_CONVS_CLS: 4 NUM_CLASSES: 2 INFERENCE_TH: 0.05 # pre_nms_thresh (default=0.05) PRE_NMS_TOP_N: 1000 # pre_nms_top_n (default=1000) NMS_TH: 0.6 # nms_thresh (default=0.6) REG_CTR_ON: True ADV: GA_DIS_LAMBDA: 0.1 # for dis loss CON_NUM_SHARED_CONV_P7: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P3: 4 # USE_DIS_GLOBAL: True USE_DIS_P7: True USE_DIS_P6: True USE_DIS_P5: True USE_DIS_P4: True USE_DIS_P3: True GRL_WEIGHT_P7: 0.02 # for gradient GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P3: 0.02 TEST: DETECTIONS_PER_IMG: 100 # fpn_post_nms_top_n (default=100) MODE: 'common' DATASETS: TRAIN_SOURCE: ("sim10k_trainval_caronly", ) TRAIN_TARGET: ("cityscapes_train_caronly_cocostyle", ) TEST: ("cityscapes_val_caronly_cocostyle", ) INPUT: MIN_SIZE_RANGE_TRAIN: (640, 800) MAX_SIZE_TRAIN: 1333 MIN_SIZE_TEST: 800 MAX_SIZE_TEST: 1333 DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: VAL_ITER: 100 ADAPT_VAL_ON: True INITIAL_AP50: 35 WEIGHT_DECAY: 0.0001 MAX_ITER: 100000 # 4 for source and 4 for target IMS_PER_BATCH: 4 CHECKPOINT_PERIOD: 100000 # BACKBONE: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" MIDDLE_HEAD: BASE_LR: 0.005 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" PLABEL_TH: (0.5, 1.0) FCOS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" # DIS: BASE_LR: 0.0025 STEPS: (90000, ) WARMUP_ITERS: 1000 WARMUP_METHOD: "constant" 2022-07-14 16:06:04,734 fcos_core INFO: Running with config: CLS_MAP_PRE: softmax DATALOADER: ASPECT_RATIO_GROUPING: True NUM_WORKERS: 4 SIZE_DIVISIBILITY: 32 DATASETS: TEST: ('cityscapes_val_caronly_cocostyle',) TRAIN_SOURCE: ('sim10k_trainval_caronly',) TRAIN_TARGET: ('cityscapes_train_caronly_cocostyle',) INPUT: MAX_SIZE_TEST: 1333 MAX_SIZE_TRAIN: 1333 MIN_SIZE_RANGE_TRAIN: (640, 800) MIN_SIZE_TEST: 800 MIN_SIZE_TRAIN: (800,) PIXEL_MEAN: [102.9801, 115.9465, 122.7717] PIXEL_STD: [1.0, 1.0, 1.0] TO_BGR255: True MODEL: ADV: BASE_DIS_TOWER: False CA_DIS_LAMBDA: 0.1 CA_DIS_P3_NUM_CONVS: 4 CA_DIS_P4_NUM_CONVS: 4 CA_DIS_P5_NUM_CONVS: 4 CA_DIS_P6_NUM_CONVS: 4 CA_DIS_P7_NUM_CONVS: 4 CA_GRL_WEIGHT_P3: 0.1 CA_GRL_WEIGHT_P4: 0.1 CA_GRL_WEIGHT_P5: 0.1 CA_GRL_WEIGHT_P6: 0.1 CA_GRL_WEIGHT_P7: 0.1 CENTER_AWARE_TYPE: ca_feature CENTER_AWARE_WEIGHT: 20 CON_DIS_LAMBDA: 0.1 CON_FUSUIN_CFG: concat CON_NUM_SHARED_CONV_P3: 4 CON_NUM_SHARED_CONV_P4: 4 CON_NUM_SHARED_CONV_P5: 4 CON_NUM_SHARED_CONV_P6: 4 CON_NUM_SHARED_CONV_P7: 4 CON_WITH_GA: False DIS_P3_NUM_CONVS: 4 DIS_P4_NUM_CONVS: 4 DIS_P5_NUM_CONVS: 4 DIS_P6_NUM_CONVS: 4 DIS_P7_NUM_CONVS: 4 GA_DIS_LAMBDA: 0.1 GRL_APPLIED_DOMAIN: both GRL_WEIGHT_P3: 0.02 GRL_WEIGHT_P4: 0.02 GRL_WEIGHT_P5: 0.02 GRL_WEIGHT_P6: 0.02 GRL_WEIGHT_P7: 0.02 OUTMAP_OP: sigmoid OUTPUT_CENTERNESS_DA: True OUTPUT_CLS_DA: True OUTPUT_REG_DA: True OUT_DIS_LAMBDA: 0.1 OUT_LOSS: ce OUT_WEIGHT: 0.5 PATCH_STRIDE: None USE_DIS_CENTER_AWARE: False USE_DIS_CON: False USE_DIS_GLOBAL: True USE_DIS_OUT: False USE_DIS_P3: True USE_DIS_P3_CON: False USE_DIS_P4: True USE_DIS_P4_CON: False USE_DIS_P5: True USE_DIS_P5_CON: False USE_DIS_P6: True USE_DIS_P6_CON: False USE_DIS_P7: True USE_DIS_P7_CON: False ATSS: ANCHOR_SIZES: (64, 128, 256, 512, 1024) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (1.0,) BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 5.0 NMS_TH: 0.6 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 POSITIVE_TYPE: ATSS PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REGRESSION_TYPE: BOX REG_LOSS_WEIGHT: 2.0 SCALES_PER_OCTAVE: 1 STRADDLE_THRESH: 0 TOPK: 9 USE_DCN_IN_TOWER: False ATSS_ON: False BACKBONE: CONV_BODY: VGG-16-FPN-RETINANET FREEZE_CONV_BODY_AT: 2 USE_GN: False VGG_W_BN: False CLS_AGNOSTIC_BBOX_REG: False DA_ON: True DEBUG_CFG: None DEVICE: cuda FBNET: ARCH: default ARCH_DEF: BN_TYPE: bn DET_HEAD_BLOCKS: [] DET_HEAD_LAST_SCALE: 1.0 DET_HEAD_STRIDE: 0 DW_CONV_SKIP_BN: True DW_CONV_SKIP_RELU: True KPTS_HEAD_BLOCKS: [] KPTS_HEAD_LAST_SCALE: 0.0 KPTS_HEAD_STRIDE: 0 MASK_HEAD_BLOCKS: [] MASK_HEAD_LAST_SCALE: 0.0 MASK_HEAD_STRIDE: 0 RPN_BN_TYPE: RPN_HEAD_BLOCKS: 0 SCALE_FACTOR: 1.0 WIDTH_DIVISOR: 1 FCOS: FPN_STRIDES: [8, 16, 32, 64, 128] INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NUM_CLASSES: 2 NUM_CONVS: 4 NUM_CONVS_CLS: 4 NUM_CONVS_REG: 4 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 REG_CTR_ON: True FCOS_ON: True FPN: USE_GN: False USE_RELU: False GROUP_NORM: DIM_PER_GP: -1 EPSILON: 1e-05 NUM_GROUPS: 32 KEYPOINT_ON: False MASK_ON: False META_ARCHITECTURE: GeneralizedRCNN MIDDLE_HEAD: ACT_LOSS: None ACT_LOSS_WEIGHT: 1.0 CAT_ACT_MAP: True CONDGRAPH_ON: True COND_WITH_BIAS: False CON_LOSS_WEIGHT: 1.0 CON_TG_CFG: KLdiv GCN1_OUT_CHANNEL: 256 GCN2_OUT_CHANNEL: 256 GCN_EDGE_NORM: softmax GCN_EDGE_PROJECT: 128 GCN_LOSS_WEIGHT: 1.0 GCN_LOSS_WEIGHT_TG: 1.0 GCN_OUT_ACTIVATION: relu GCN_SHORTCUT: False GLOBAL_GRAPH_ON: False GM: BG_RATIO: 2 MATCHING_CFG: o2o MATCHING_LOSS_CFG: MSE MATCHING_LOSS_WEIGHT: 1.0 MIN_AP50_SAVE: 40 NODE_DIS_LAMBDA: 0.02 NODE_DIS_PLACE: feat NODE_DIS_WEIGHT: 0.1 NODE_LOSS_WEIGHT: 1.0 NUM_NODES_PER_LVL_SR: 100 NUM_NODES_PER_LVL_TG: 100 WITH_CLUSTER_UPDATE: True WITH_COMPLETE_GRAPH: True WITH_COND_CLS: False WITH_CTR: False WITH_DOMAIN_INTERACTION: True WITH_GLOBAL_GRAPH: False WITH_NODE_DIS: True WITH_QUADRATIC_MATCHING: True WITH_SCORE_WEIGHT: False WITH_SEMANTIC_COMPLETION: True GM_ON: False IN_NORM: LN NUM_CONVS_IN: 2 NUM_CONVS_OUT: 1 PROTO_CHANNEL: 256 PROTO_MOMENTUM: 0.95 PROTO_WITH_BG: True RETURN_ACT_LOGITS: False MIDDLE_HEAD_CFG: GM_HEAD RESNETS: BACKBONE_OUT_CHANNELS: 1024 NUM_GROUPS: 1 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_FUNC: StemWithFixedBatchNorm STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: True TRANS_FUNC: BottleneckWithFixedBatchNorm WIDTH_PER_GROUP: 64 RETINANET: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDES: (8, 16, 32, 64, 128) ASPECT_RATIOS: (0.5, 1.0, 2.0) BBOX_REG_BETA: 0.11 BBOX_REG_WEIGHT: 4.0 BG_IOU_THRESHOLD: 0.4 FG_IOU_THRESHOLD: 0.5 INFERENCE_TH: 0.05 LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.4 NUM_CLASSES: 81 NUM_CONVS: 4 OCTAVE: 2.0 PRE_NMS_TOP_N: 1000 PRIOR_PROB: 0.01 SCALES_PER_OCTAVE: 3 STRADDLE_THRESH: 0 USE_C5: False RETINANET_ON: False ROI_BOX_HEAD: CONV_HEAD_DIM: 256 DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 81 NUM_STACKED_CONVS: 4 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: FastRCNNPredictor USE_GN: False ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0) BG_IOU_THRESHOLD: 0.5 DETECTIONS_PER_IMG: 100 FG_IOU_THRESHOLD: 0.5 NMS: 0.5 POSITIVE_FRACTION: 0.25 SCORE_THRESH: 0.05 USE_FPN: False ROI_KEYPOINT_HEAD: CONV_LAYERS: (512, 512, 512, 512, 512, 512, 512, 512) FEATURE_EXTRACTOR: KeypointRCNNFeatureExtractor MLP_HEAD_DIM: 1024 NUM_CLASSES: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) PREDICTOR: KeypointRCNNPredictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True ROI_MASK_HEAD: CONV_LAYERS: (256, 256, 256, 256) DILATION: 1 FEATURE_EXTRACTOR: ResNet50Conv5ROIFeatureExtractor MLP_HEAD_DIM: 1024 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_SCALES: (0.0625,) POSTPROCESS_MASKS: False POSTPROCESS_MASKS_THRESHOLD: 0.5 PREDICTOR: MaskRCNNC4Predictor RESOLUTION: 14 SHARE_BOX_FEATURE_EXTRACTOR: True USE_GN: False RPN: ANCHOR_SIZES: (32, 64, 128, 256, 512) ANCHOR_STRIDE: (16,) ASPECT_RATIOS: (0.5, 1.0, 2.0) BATCH_SIZE_PER_IMAGE: 256 BG_IOU_THRESHOLD: 0.3 FG_IOU_THRESHOLD: 0.7 FPN_POST_NMS_TOP_N_TEST: 2000 FPN_POST_NMS_TOP_N_TRAIN: 2000 MIN_SIZE: 0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 6000 PRE_NMS_TOP_N_TRAIN: 12000 RPN_HEAD: SingleConvRPNHead STRADDLE_THRESH: 0 USE_FPN: False RPN_ONLY: True USE_SYNCBN: False WEIGHT: https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth OUTPUT_DIR: ./experiments/sigma/sim10k_to_cityscapes_vgg16/ PATHS_CATALOG: /home/sh/SIGMA-main/fcos_core/config/paths_catalog.py SOLVER: ADAPT_VAL_ON: True BACKBONE: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) SWA: False WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant CHECKPOINT_PERIOD: 100000 DIS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant FCOS: BASE_LR: 0.0025 BIAS_LR_FACTOR: 2 GAMMA: 0.1 STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant IMS_PER_BATCH: 4 INITIAL_AP50: 35 MAX_ITER: 100000 MIDDLE_HEAD: BASE_LR: 0.005 BIAS_LR_FACTOR: 2 GAMMA: 0.1 PLABEL_TH: (0.5, 1.0) STEPS: (90000,) WARMUP_FACTOR: 0.3333333333333333 WARMUP_ITERS: 1000 WARMUP_METHOD: constant MOMENTUM: 0.9 VAL_ITER: 100 VAL_TYPE: AP50 WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0 TENSORBOARD_EXPERIMENT: ./exps/demo/logs/ TEST: DETECTIONS_PER_IMG: 100 EXPECTED_RESULTS: [] EXPECTED_RESULTS_SIGMA_TOL: 4 IMS_PER_BATCH: 4 MODE: common 2022-07-14 16:06:11,212 fcos_core.trainer INFO: node dis setting: feat 2022-07-14 16:06:11,256 fcos_core.trainer INFO: node_dis initialized 2022-07-14 16:06:11,258 fcos_core.trainer INFO: node_cls_middle initialized 2022-07-14 16:06:11,261 fcos_core.trainer INFO: head_in_ln initialized 2022-07-14 16:06:11,665 fcos_core.utils.checkpoint INFO: Loading checkpoint from https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth 2022-07-14 16:06:11,665 fcos_core.utils.checkpoint INFO: url https://s3.ap-northeast-2.amazonaws.com/open-mmlab/pretrain/third_party/vgg16_caffe-292e1171.pth cached in /home/sh/.torch/models/vgg16_caffe-292e1171.pth 2022-07-14 16:06:11,738 fcos_core.utils.model_serialization INFO: body.features.0.bias loaded from features.0.bias of shape (64,) 2022-07-14 16:06:11,739 fcos_core.utils.model_serialization INFO: body.features.0.weight loaded from features.0.weight of shape (64, 3, 3, 3) 2022-07-14 16:06:11,739 fcos_core.utils.model_serialization INFO: body.features.10.bias loaded from features.10.bias of shape (256,) 2022-07-14 16:06:11,739 fcos_core.utils.model_serialization INFO: body.features.10.weight loaded from features.10.weight of shape (256, 128, 3, 3) 2022-07-14 16:06:11,739 fcos_core.utils.model_serialization INFO: body.features.12.bias loaded from features.12.bias of shape (256,) 2022-07-14 16:06:11,739 fcos_core.utils.model_serialization INFO: body.features.12.weight loaded from features.12.weight of shape (256, 256, 3, 3) 2022-07-14 16:06:11,739 fcos_core.utils.model_serialization INFO: body.features.14.bias loaded from features.14.bias of shape (256,) 2022-07-14 16:06:11,739 fcos_core.utils.model_serialization INFO: body.features.14.weight loaded from features.14.weight of shape (256, 256, 3, 3) 2022-07-14 16:06:11,739 fcos_core.utils.model_serialization INFO: body.features.17.bias loaded from features.17.bias of shape (512,) 2022-07-14 16:06:11,740 fcos_core.utils.model_serialization INFO: body.features.17.weight loaded from features.17.weight of shape (512, 256, 3, 3) 2022-07-14 16:06:11,740 fcos_core.utils.model_serialization INFO: body.features.19.bias loaded from features.19.bias of shape (512,) 2022-07-14 16:06:11,740 fcos_core.utils.model_serialization INFO: body.features.19.weight loaded from features.19.weight of shape (512, 512, 3, 3) 2022-07-14 16:06:11,740 fcos_core.utils.model_serialization INFO: body.features.2.bias loaded from features.2.bias of shape (64,) 2022-07-14 16:06:11,740 fcos_core.utils.model_serialization INFO: body.features.2.weight loaded from features.2.weight of shape (64, 64, 3, 3) 2022-07-14 16:06:11,740 fcos_core.utils.model_serialization INFO: body.features.21.bias loaded from features.21.bias of shape (512,) 2022-07-14 16:06:11,740 fcos_core.utils.model_serialization INFO: body.features.21.weight loaded from features.21.weight of shape (512, 512, 3, 3) 2022-07-14 16:06:11,740 fcos_core.utils.model_serialization INFO: body.features.24.bias loaded from features.24.bias of shape (512,) 2022-07-14 16:06:11,740 fcos_core.utils.model_serialization INFO: body.features.24.weight loaded from features.24.weight of shape (512, 512, 3, 3) 2022-07-14 16:06:11,741 fcos_core.utils.model_serialization INFO: body.features.26.bias loaded from features.26.bias of shape (512,) 2022-07-14 16:06:11,741 fcos_core.utils.model_serialization INFO: body.features.26.weight loaded from features.26.weight of shape (512, 512, 3, 3) 2022-07-14 16:06:11,741 fcos_core.utils.model_serialization INFO: body.features.28.bias loaded from features.28.bias of shape (512,) 2022-07-14 16:06:11,741 fcos_core.utils.model_serialization INFO: body.features.28.weight loaded from features.28.weight of shape (512, 512, 3, 3) 2022-07-14 16:06:11,741 fcos_core.utils.model_serialization INFO: body.features.5.bias loaded from features.5.bias of shape (128,) 2022-07-14 16:06:11,741 fcos_core.utils.model_serialization INFO: body.features.5.weight loaded from features.5.weight of shape (128, 64, 3, 3) 2022-07-14 16:06:11,741 fcos_core.utils.model_serialization INFO: body.features.7.bias loaded from features.7.bias of shape (128,) 2022-07-14 16:06:11,741 fcos_core.utils.model_serialization INFO: body.features.7.weight loaded from features.7.weight of shape (128, 128, 3, 3) 2022-07-14 16:06:13,113 fcos_core.data.build WARNING: When using more than one image per GPU you may encounter an out-of-memory (OOM) error if your GPU does not have sufficient memory. If this happens, you can reduce SOLVER.IMS_PER_BATCH (for training) or TEST.IMS_PER_BATCH (for inference). For training, you must also adjust the learning rate and schedule length according to the linear scaling rule. See for example: https://github.com/facebookresearch/Detectron/blob/master/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14 2022-07-14 16:06:17,752 fcos_core.trainer INFO: Start training 2022-07-14 16:07:17,072 fcos_core.trainer INFO: eta: 3 days, 10:21:57 iter: 20 loss_ds: 5.5951 (5.9816) node_loss: 0.6647 (0.6668) mat_loss_aff: 0.9909 (0.9903) mat_loss_qu: 0.0005 (0.0006) loss_cls: 0.6391 (0.7085) loss_reg: 1.8815 (2.2282) loss_centerness: 0.6642 (0.6733) loss_adv_P7: 0.1400 (0.1400) loss_adv_P6: 0.1395 (0.1397) loss_adv_P5: 0.1404 (0.1406) loss_adv_P4: 0.1372 (0.1375) loss_adv_P3: 0.1383 (0.1382) time: 2.6906 (2.9658) data: 0.0167 (0.1196) dis_loss: 0.0719 (0.0715) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 7115 2022-07-14 16:08:10,735 fcos_core.trainer INFO: eta: 3 days, 6:25:31 iter: 40 loss_ds: 4.6222 (5.3299) node_loss: 0.6231 (0.6464) mat_loss_aff: 0.9898 (0.9901) mat_loss_qu: 0.0006 (0.0006) loss_cls: 0.5555 (0.6438) loss_reg: 1.0923 (1.6839) loss_centerness: 0.6571 (0.6668) loss_adv_P7: 0.1370 (0.1386) loss_adv_P6: 0.1359 (0.1378) loss_adv_P5: 0.1364 (0.1382) loss_adv_P4: 0.1314 (0.1341) loss_adv_P3: 0.1287 (0.1332) time: 2.5548 (2.8244) data: 0.0212 (0.0757) dis_loss: 0.0595 (0.0602) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 7115 2022-07-14 16:09:06,698 fcos_core.trainer INFO: eta: 3 days, 6:10:02 iter: 60 loss_ds: 4.1711 (4.9577) node_loss: 0.5752 (0.6244) mat_loss_aff: 0.9942 (0.9910) mat_loss_qu: 0.0006 (0.0006) loss_cls: 0.3578 (0.5552) loss_reg: 0.9743 (1.4405) loss_centerness: 0.6571 (0.6633) loss_adv_P7: 0.1331 (0.1368) loss_adv_P6: 0.1278 (0.1347) loss_adv_P5: 0.1276 (0.1346) loss_adv_P4: 0.1205 (0.1291) loss_adv_P3: 0.1105 (0.1258) time: 2.8541 (2.8157) data: 0.0177 (0.0600) dis_loss: 0.0430 (0.0501) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:10:02,429 fcos_core.trainer INFO: eta: 3 days, 5:56:56 iter: 80 loss_ds: 3.8044 (4.6846) node_loss: 0.4102 (0.5727) mat_loss_aff: 0.9906 (0.9905) mat_loss_qu: 0.0006 (0.0006) loss_cls: 0.2973 (0.4983) loss_reg: 0.8808 (1.3088) loss_centerness: 0.6543 (0.6613) loss_adv_P7: 0.1224 (0.1329) loss_adv_P6: 0.1153 (0.1292) loss_adv_P5: 0.1143 (0.1289) loss_adv_P4: 0.1010 (0.1225) loss_adv_P3: 0.0828 (0.1155) time: 2.9220 (2.8084) data: 0.0222 (0.0539) dis_loss: 0.0307 (0.0460) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:10:59,864 fcos_core.trainer INFO: eta: 3 days, 6:17:05 iter: 100 loss_ds: 3.5474 (4.4552) node_loss: 0.2654 (0.5122) mat_loss_aff: 0.9883 (0.9901) mat_loss_qu: 0.0005 (0.0006) loss_cls: 0.3316 (0.4681) loss_reg: 0.8163 (1.2149) loss_centerness: 0.6458 (0.6581) loss_adv_P7: 0.1015 (0.1261) loss_adv_P6: 0.0968 (0.1227) loss_adv_P5: 0.0993 (0.1227) loss_adv_P4: 0.0757 (0.1130) loss_adv_P3: 0.0546 (0.1034) time: 2.6825 (2.8211) data: 0.0203 (0.0491) dis_loss: 0.0299 (0.0447) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:10:59,865 fcos_core.inference INFO: Start evaluation on ('cityscapes_val_caronly_cocostyle',) dataset(500 images). 2022-07-14 16:11:34,785 fcos_core.inference INFO: Preparing results for COCO format 2022-07-14 16:11:34,786 fcos_core.inference INFO: Preparing bbox results 2022-07-14 16:11:35,405 fcos_core.inference INFO: Evaluating predictions 2022-07-14 16:11:56,153 fcos_core.inference INFO: OrderedDict([('bbox', OrderedDict([('AP', 0.023905415169108813), ('AP50', 0.10343661571290853), ('AP75', 0.0017003897700123725), ('APs', 0.005940594059405941), ('APm', 0.037964234636442204), ('APl', 0.05042234045187016)]))]) 2022-07-14 16:12:55,675 fcos_core.trainer INFO: eta: 3 days, 20:00:00 iter: 120 loss_ds: 3.2271 (4.2557) node_loss: 0.2042 (0.4617) mat_loss_aff: 0.9916 (0.9900) mat_loss_qu: 0.0005 (0.0006) loss_cls: 0.3245 (0.4428) loss_reg: 0.7835 (1.1440) loss_centerness: 0.6378 (0.6548) loss_adv_P7: 0.0644 (0.1163) loss_adv_P6: 0.0674 (0.1146) loss_adv_P5: 0.0664 (0.1143) loss_adv_P4: 0.0428 (0.1017) loss_adv_P3: 0.0302 (0.0915) time: 2.8290 (3.3160) data: 0.0268 (0.5161) dis_loss: 0.0268 (0.0426) AP: 2.3905 (2.3905) AP50: 10.3437 (10.3437) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:13:50,689 fcos_core.trainer INFO: eta: 3 days, 17:44:30 iter: 140 loss_ds: 2.9930 (4.0810) node_loss: 0.1749 (0.4221) mat_loss_aff: 0.9921 (0.9901) mat_loss_qu: 0.0005 (0.0006) loss_cls: 0.2784 (0.4201) loss_reg: 0.6963 (1.0850) loss_centerness: 0.6325 (0.6520) loss_adv_P7: 0.0395 (0.1059) loss_adv_P6: 0.0449 (0.1050) loss_adv_P5: 0.0393 (0.1042) loss_adv_P4: 0.0250 (0.0909) loss_adv_P3: 0.0158 (0.0809) time: 2.5997 (3.2352) data: 0.0261 (0.4471) dis_loss: 0.0269 (0.0413) AP: 2.3905 (2.3905) AP50: 10.3437 (10.3437) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:14:45,041 fcos_core.trainer INFO: eta: 3 days, 15:55:45 iter: 160 loss_ds: 2.9898 (3.9419) node_loss: 0.1577 (0.3910) mat_loss_aff: 0.9925 (0.9903) mat_loss_qu: 0.0005 (0.0006) loss_cls: 0.2570 (0.4003) loss_reg: 0.7111 (1.0392) loss_centerness: 0.6336 (0.6495) loss_adv_P7: 0.0385 (0.0983) loss_adv_P6: 0.0389 (0.0974) loss_adv_P5: 0.0395 (0.0970) loss_adv_P4: 0.0191 (0.0824) loss_adv_P3: 0.0088 (0.0722) time: 2.6543 (3.1705) data: 0.0204 (0.3952) dis_loss: 0.0234 (0.0385) AP: 2.3905 (2.3905) AP50: 10.3437 (10.3437) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:15:39,971 fcos_core.trainer INFO: eta: 3 days, 14:36:19 iter: 180 loss_ds: 2.8606 (3.8230) node_loss: 0.1312 (0.3633) mat_loss_aff: 0.9900 (0.9901) mat_loss_qu: 0.0005 (0.0006) loss_cls: 0.2490 (0.3859) loss_reg: 0.6949 (1.0031) loss_centerness: 0.6185 (0.6465) loss_adv_P7: 0.0281 (0.0907) loss_adv_P6: 0.0261 (0.0897) loss_adv_P5: 0.0226 (0.0891) loss_adv_P4: 0.0077 (0.0743) loss_adv_P3: 0.0059 (0.0649) time: 2.7971 (3.1234) data: 0.0242 (0.3547) dis_loss: 0.0234 (0.0385) AP: 2.3905 (2.3905) AP50: 10.3437 (10.3437) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:16:32,294 fcos_core.trainer INFO: eta: 3 days, 13:10:54 iter: 200 loss_ds: 2.8660 (3.7268) node_loss: 0.1446 (0.3416) mat_loss_aff: 0.9889 (0.9900) mat_loss_qu: 0.0005 (0.0006) loss_cls: 0.2585 (0.3739) loss_reg: 0.6975 (0.9734) loss_centerness: 0.6217 (0.6437) loss_adv_P7: 0.0199 (0.0840) loss_adv_P6: 0.0164 (0.0835) loss_adv_P5: 0.0172 (0.0831) loss_adv_P4: 0.0080 (0.0677) loss_adv_P3: 0.0043 (0.0589) time: 2.5810 (3.0727) data: 0.0179 (0.3227) dis_loss: 0.0380 (0.0395) AP: 2.3905 (2.3905) AP50: 10.3437 (10.3437) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:16:32,294 fcos_core.inference INFO: Start evaluation on ('cityscapes_val_caronly_cocostyle',) dataset(500 images). 2022-07-14 16:17:07,228 fcos_core.inference INFO: Preparing results for COCO format 2022-07-14 16:17:07,229 fcos_core.inference INFO: Preparing bbox results 2022-07-14 16:17:07,872 fcos_core.inference INFO: Evaluating predictions 2022-07-14 16:17:29,186 fcos_core.inference INFO: OrderedDict([('bbox', OrderedDict([('AP', 0.049269951610873795), ('AP50', 0.18966787518735956), ('AP75', 0.007054427399597016), ('APs', 0.012735156815480342), ('APm', 0.08565419375918452), ('APl', 0.07309296179115009)]))]) 2022-07-14 16:18:19,795 fcos_core.trainer INFO: eta: 3 days, 18:57:57 iter: 220 loss_ds: 2.7394 (3.6444) node_loss: 0.1407 (0.3243) mat_loss_aff: 0.9885 (0.9898) mat_loss_qu: 0.0006 (0.0006) loss_cls: 0.2576 (0.3632) loss_reg: 0.6513 (0.9467) loss_centerness: 0.6143 (0.6410) loss_adv_P7: 0.0165 (0.0789) loss_adv_P6: 0.0191 (0.0788) loss_adv_P5: 0.0209 (0.0785) loss_adv_P4: 0.0076 (0.0630) loss_adv_P3: 0.0034 (0.0540) time: 2.4537 (3.2820) data: 0.0210 (0.5552) dis_loss: 0.0217 (0.0378) AP: 2.3905 (3.6588) AP50: 10.3437 (14.6552) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:19:14,289 fcos_core.trainer INFO: eta: 3 days, 17:39:38 iter: 240 loss_ds: 2.7836 (3.5791) node_loss: 0.1281 (0.3094) mat_loss_aff: 0.9881 (0.9896) mat_loss_qu: 0.0006 (0.0006) loss_cls: 0.2451 (0.3546) loss_reg: 0.6987 (0.9279) loss_centerness: 0.6156 (0.6393) loss_adv_P7: 0.0156 (0.0746) loss_adv_P6: 0.0193 (0.0747) loss_adv_P5: 0.0202 (0.0746) loss_adv_P4: 0.0085 (0.0585) loss_adv_P3: 0.0028 (0.0497) time: 2.5006 (3.2355) data: 0.0262 (0.5121) dis_loss: 0.0308 (0.0377) AP: 2.3905 (3.6588) AP50: 10.3437 (14.6552) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:20:11,308 fcos_core.trainer INFO: eta: 3 days, 16:49:23 iter: 260 loss_ds: 2.7274 (3.5145) node_loss: 0.1368 (0.2970) mat_loss_aff: 0.9900 (0.9894) mat_loss_qu: 0.0005 (0.0006) loss_cls: 0.2411 (0.3465) loss_reg: 0.6747 (0.9073) loss_centerness: 0.6167 (0.6377) loss_adv_P7: 0.0124 (0.0700) loss_adv_P6: 0.0159 (0.0704) loss_adv_P5: 0.0118 (0.0700) loss_adv_P4: 0.0037 (0.0544) loss_adv_P3: 0.0026 (0.0461) time: 2.7891 (3.2060) data: 0.0206 (0.4752) dis_loss: 0.0210 (0.0368) AP: 2.3905 (3.6588) AP50: 10.3437 (14.6552) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:21:02,382 fcos_core.trainer INFO: eta: 3 days, 15:30:53 iter: 280 loss_ds: 2.6770 (3.4574) node_loss: 0.1272 (0.2859) mat_loss_aff: 0.9898 (0.9894) mat_loss_qu: 0.0006 (0.0006) loss_cls: 0.2280 (0.3385) loss_reg: 0.6004 (0.8872) loss_centerness: 0.6100 (0.6359) loss_adv_P7: 0.0155 (0.0665) loss_adv_P6: 0.0174 (0.0674) loss_adv_P5: 0.0192 (0.0668) loss_adv_P4: 0.0041 (0.0509) loss_adv_P3: 0.0025 (0.0430) time: 2.5606 (3.1594) data: 0.0177 (0.4433) dis_loss: 0.0231 (0.0361) AP: 2.3905 (3.6588) AP50: 10.3437 (14.6552) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:21:58,270 fcos_core.trainer INFO: eta: 3 days, 14:49:24 iter: 300 loss_ds: 2.6267 (3.4052) node_loss: 0.1203 (0.2758) mat_loss_aff: 0.9871 (0.9892) mat_loss_qu: 0.0006 (0.0006) loss_cls: 0.2158 (0.3312) loss_reg: 0.5932 (0.8685) loss_centerness: 0.6092 (0.6345) loss_adv_P7: 0.0141 (0.0632) loss_adv_P6: 0.0149 (0.0645) loss_adv_P5: 0.0155 (0.0638) loss_adv_P4: 0.0032 (0.0479) loss_adv_P3: 0.0020 (0.0404) time: 2.7953 (3.1350) data: 0.0201 (0.4159) dis_loss: 0.0286 (0.0359) AP: 2.3905 (3.6588) AP50: 10.3437 (14.6552) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:21:58,271 fcos_core.inference INFO: Start evaluation on ('cityscapes_val_caronly_cocostyle',) dataset(500 images). 2022-07-14 16:22:32,944 fcos_core.inference INFO: Preparing results for COCO format 2022-07-14 16:22:32,945 fcos_core.inference INFO: Preparing bbox results 2022-07-14 16:22:33,507 fcos_core.inference INFO: Evaluating predictions 2022-07-14 16:22:53,924 fcos_core.inference INFO: OrderedDict([('bbox', OrderedDict([('AP', 0.07319755417807303), ('AP50', 0.25468018233421696), ('AP75', 0.016027832753233152), ('APs', 0.02230816372555857), ('APm', 0.08539146520089637), ('APl', 0.14876022216250454)]))]) 2022-07-14 16:23:46,598 fcos_core.trainer INFO: eta: 3 days, 18:45:13 iter: 320 loss_ds: 2.6093 (3.3559) node_loss: 0.1101 (0.2660) mat_loss_aff: 0.9861 (0.9890) mat_loss_qu: 0.0007 (0.0006) loss_cls: 0.2185 (0.3246) loss_reg: 0.6034 (0.8517) loss_centerness: 0.6097 (0.6330) loss_adv_P7: 0.0081 (0.0598) loss_adv_P6: 0.0113 (0.0614) loss_adv_P5: 0.0119 (0.0608) loss_adv_P4: 0.0022 (0.0451) loss_adv_P3: 0.0015 (0.0380) time: 2.5850 (3.2776) data: 0.0221 (0.5660) dis_loss: 0.0209 (0.0354) AP: 4.9270 (4.8791) AP50: 18.9668 (18.2595) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:24:41,196 fcos_core.trainer INFO: eta: 3 days, 17:50:37 iter: 340 loss_ds: 2.6602 (3.3169) node_loss: 0.1218 (0.2585) mat_loss_aff: 0.9869 (0.9889) mat_loss_qu: 0.0006 (0.0006) loss_cls: 0.2156 (0.3181) loss_reg: 0.6456 (0.8397) loss_centerness: 0.6067 (0.6316) loss_adv_P7: 0.0098 (0.0569) loss_adv_P6: 0.0171 (0.0595) loss_adv_P5: 0.0132 (0.0586) loss_adv_P4: 0.0022 (0.0426) loss_adv_P3: 0.0013 (0.0358) time: 2.6529 (3.2454) data: 0.0426 (0.5348) dis_loss: 0.0252 (0.0350) AP: 4.9270 (4.8791) AP50: 18.9668 (18.2595) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:25:38,033 fcos_core.trainer INFO: eta: 3 days, 17:12:18 iter: 360 loss_ds: 2.5523 (3.2756) node_loss: 0.1284 (0.2516) mat_loss_aff: 0.9894 (0.9888) mat_loss_qu: 0.0008 (0.0006) loss_cls: 0.2169 (0.3130) loss_reg: 0.5290 (0.8225) loss_centerness: 0.6052 (0.6301) loss_adv_P7: 0.0091 (0.0543) loss_adv_P6: 0.0213 (0.0577) loss_adv_P5: 0.0158 (0.0565) loss_adv_P4: 0.0036 (0.0405) loss_adv_P3: 0.0020 (0.0340) time: 2.7810 (3.2230) data: 0.0251 (0.5068) dis_loss: 0.0160 (0.0343) AP: 4.9270 (4.8791) AP50: 18.9668 (18.2595) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:26:33,419 fcos_core.trainer INFO: eta: 3 days, 16:31:35 iter: 380 loss_ds: 2.5700 (3.2386) node_loss: 0.1251 (0.2450) mat_loss_aff: 0.9851 (0.9885) mat_loss_qu: 0.0007 (0.0006) loss_cls: 0.2327 (0.3087) loss_reg: 0.5308 (0.8071) loss_centerness: 0.6043 (0.6287) loss_adv_P7: 0.0087 (0.0534) loss_adv_P6: 0.0107 (0.0560) loss_adv_P5: 0.0051 (0.0539) loss_adv_P4: 0.0019 (0.0386) loss_adv_P3: 0.0013 (0.0323) time: 2.7817 (3.1991) data: 0.0288 (0.4818) dis_loss: 0.0228 (0.0340) AP: 4.9270 (4.8791) AP50: 18.9668 (18.2595) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:27:32,902 fcos_core.trainer INFO: eta: 3 days, 16:11:51 iter: 400 loss_ds: 2.5207 (3.2032) node_loss: 0.1026 (0.2377) mat_loss_aff: 0.9881 (0.9884) mat_loss_qu: 0.0006 (0.0006) loss_cls: 0.2068 (0.3041) loss_reg: 0.5465 (0.7940) loss_centerness: 0.6031 (0.6274) loss_adv_P7: 0.0067 (0.0513) loss_adv_P6: 0.0103 (0.0541) loss_adv_P5: 0.0050 (0.0516) loss_adv_P4: 0.0015 (0.0368) loss_adv_P3: 0.0010 (0.0307) time: 2.9404 (3.1879) data: 0.0258 (0.4594) dis_loss: 0.0197 (0.0343) AP: 4.9270 (4.8791) AP50: 18.9668 (18.2595) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:27:32,903 fcos_core.inference INFO: Start evaluation on ('cityscapes_val_caronly_cocostyle',) dataset(500 images). 2022-07-14 16:28:09,200 fcos_core.inference INFO: Preparing results for COCO format 2022-07-14 16:28:09,200 fcos_core.inference INFO: Preparing bbox results 2022-07-14 16:28:10,278 fcos_core.inference INFO: Evaluating predictions 2022-07-14 16:28:31,014 fcos_core.inference INFO: OrderedDict([('bbox', OrderedDict([('AP', 0.07242469928647614), ('AP50', 0.25218463495056187), ('AP75', 0.00777119359051473), ('APs', 0.021353417247329226), ('APm', 0.09877233598066563), ('APl', 0.1337989383399907)]))]) 2022-07-14 16:29:30,240 fcos_core.trainer INFO: eta: 3 days, 19:42:31 iter: 420 loss_ds: 2.6287 (3.1772) node_loss: 0.1369 (0.2335) mat_loss_aff: 0.9880 (0.9883) mat_loss_qu: 0.0007 (0.0006) loss_cls: 0.2251 (0.3003) loss_reg: 0.5702 (0.7850) loss_centerness: 0.6072 (0.6265) loss_adv_P7: 0.0076 (0.0495) loss_adv_P6: 0.0141 (0.0529) loss_adv_P5: 0.0096 (0.0499) loss_adv_P4: 0.0014 (0.0351) loss_adv_P3: 0.0010 (0.0293) time: 2.8450 (3.3154) data: 0.0180 (0.5775) dis_loss: 0.0239 (0.0338) AP: 4.9270 (5.4699) AP50: 18.9668 (19.9992) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:30:31,135 fcos_core.trainer INFO: eta: 3 days, 19:20:59 iter: 440 loss_ds: 2.5938 (3.1492) node_loss: 0.1138 (0.2283) mat_loss_aff: 0.9851 (0.9882) mat_loss_qu: 0.0007 (0.0006) loss_cls: 0.2263 (0.2972) loss_reg: 0.5482 (0.7740) loss_centerness: 0.6080 (0.6256) loss_adv_P7: 0.0048 (0.0475) loss_adv_P6: 0.0104 (0.0515) loss_adv_P5: 0.0061 (0.0482) loss_adv_P4: 0.0012 (0.0336) loss_adv_P3: 0.0010 (0.0281) time: 2.9607 (3.3031) data: 0.0184 (0.5527) dis_loss: 0.0235 (0.0338) AP: 4.9270 (5.4699) AP50: 18.9668 (19.9992) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:31:33,868 fcos_core.trainer INFO: eta: 3 days, 19:07:53 iter: 460 loss_ds: 2.6399 (3.1257) node_loss: 0.1238 (0.2243) mat_loss_aff: 0.9846 (0.9879) mat_loss_qu: 0.0007 (0.0006) loss_cls: 0.2368 (0.2942) loss_reg: 0.5267 (0.7634) loss_centerness: 0.5989 (0.6246) loss_adv_P7: 0.0078 (0.0458) loss_adv_P6: 0.0353 (0.0520) loss_adv_P5: 0.0110 (0.0469) loss_adv_P4: 0.0016 (0.0322) loss_adv_P3: 0.0011 (0.0269) time: 3.2144 (3.2959) data: 0.0350 (0.5303) dis_loss: 0.0306 (0.0338) AP: 4.9270 (5.4699) AP50: 18.9668 (19.9992) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:32:34,888 fcos_core.trainer INFO: eta: 3 days, 18:49:52 iter: 480 loss_ds: 2.6293 (3.1074) node_loss: 0.1143 (0.2202) mat_loss_aff: 0.9823 (0.9877) mat_loss_qu: 0.0006 (0.0007) loss_cls: 0.2086 (0.2911) loss_reg: 0.5243 (0.7543) loss_centerness: 0.6065 (0.6239) loss_adv_P7: 0.0166 (0.0451) loss_adv_P6: 0.0562 (0.0533) loss_adv_P5: 0.0215 (0.0464) loss_adv_P4: 0.0023 (0.0314) loss_adv_P3: 0.0018 (0.0263) time: 2.9104 (3.2857) data: 0.0260 (0.5097) dis_loss: 0.0369 (0.0340) AP: 4.9270 (5.4699) AP50: 18.9668 (19.9992) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:33:29,720 fcos_core.trainer INFO: eta: 3 days, 18:12:40 iter: 500 loss_ds: 2.5630 (3.0881) node_loss: 0.1206 (0.2164) mat_loss_aff: 0.9822 (0.9873) mat_loss_qu: 0.0007 (0.0007) loss_cls: 0.2414 (0.2893) loss_reg: 0.4995 (0.7443) loss_centerness: 0.6054 (0.6231) loss_adv_P7: 0.0061 (0.0436) loss_adv_P6: 0.0399 (0.0531) loss_adv_P5: 0.0472 (0.0476) loss_adv_P4: 0.0015 (0.0303) loss_adv_P3: 0.0009 (0.0253) time: 2.6623 (3.2639) data: 0.0407 (0.4907) dis_loss: 0.0274 (0.0339) AP: 4.9270 (5.4699) AP50: 18.9668 (19.9992) lr_backbone: 0.000833 lr_middle_head: 0.001667 lr_fcos: 0.000833 lr_dis: 0.000833 max mem: 8023 2022-07-14 16:33:29,720 fcos_core.inference INFO: Start evaluation on ('cityscapes_val_caronly_cocostyle',) dataset(500 images). 2022-07-14 16:34:04,748 fcos_core.inference INFO: Preparing results for COCO format 2022-07-14 16:34:04,748 fcos_core.inference INFO: Preparing bbox results 2022-07-14 16:34:05,343 fcos_core.inference INFO: Evaluating predictions 2022-07-14 16:34:26,438 fcos_core.inference INFO: OrderedDict([('bbox', OrderedDict([('AP', 0.11023148438641284), ('AP50', 0.2934023642821854), ('AP75', 0.05630611467158457), ('APs', 0.022213713521963545), ('APm', 0.13591090929229557), ('APl', 0.223972915178203)]))])