voxelnext_ioubranch_large.yaml

CLASS_NAMES: ['Vehicle', 'Pedestrian', 'Cyclist']

DATA_CONFIG:
    _BASE_CONFIG_: cfgs/dataset_configs/waymo_dataset.yaml

MODEL:
    NAME: VoxelNeXt

    VFE:
        NAME: MeanVFE

    BACKBONE_3D:
        NAME: VoxelResBackBone8xVoxelNeXt
        SPCONV_KERNEL_SIZES: [5, 5, 3, 3]
        OUT_CHANNEL: 256
        CHANNELS: [32, 64, 128, 256, 256]

    DENSE_HEAD:
        NAME: VoxelNeXtHead
        IOU_BRANCH: True
        CLASS_AGNOSTIC: False
        INPUT_FEATURES: 256

        CLASS_NAMES_EACH_HEAD: [
            ['Vehicle', 'Pedestrian', 'Cyclist']
        ]

        SHARED_CONV_CHANNEL: 256
        USE_BIAS_BEFORE_NORM: True
        NUM_HM_CONV: 2
        SEPARATE_HEAD_CFG:
            HEAD_ORDER: ['center', 'center_z', 'dim', 'rot']
            HEAD_DICT: {
                'center': {'out_channels': 2, 'num_conv': 2},
                'center_z': {'out_channels': 1, 'num_conv': 2},
                'dim': {'out_channels': 3, 'num_conv': 2},
                'rot': {'out_channels': 2, 'num_conv': 2},
                'iou': {'out_channels': 1, 'num_conv': 2},
            }
        RECTIFIER: [0.68, 0.71, 0.65]
        TARGET_ASSIGNER_CONFIG:
            FEATURE_MAP_STRIDE: 8
            NUM_MAX_OBJS: 500
            GAUSSIAN_OVERLAP: 0.1
            MIN_RADIUS: 2

        LOSS_CONFIG:
            LOSS_WEIGHTS: {
                'cls_weight': 1.0,
                'loc_weight': 2.0,
                'code_weights': [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
            }

        POST_PROCESSING:
            SCORE_THRESH: 0.1
            POST_CENTER_LIMIT_RANGE: [-75.2, -75.2, -2, 75.2, 75.2, 4]
            MAX_OBJ_PER_SAMPLE: 500
            NMS_CONFIG:
                NMS_TYPE: nms_gpu
                NMS_THRESH: [0.8, 0.55, 0.55] #0.7
                NMS_PRE_MAXSIZE: [2048, 1024, 1024] #[4096]
                NMS_POST_MAXSIZE: [200, 150, 150] #500

    POST_PROCESSING:
        RECALL_THRESH_LIST: [0.3, 0.5, 0.7]

        EVAL_METRIC: waymo


OPTIMIZATION:
    BATCH_SIZE_PER_GPU: 4
    NUM_EPOCHS: 12

    OPTIMIZER: adam_onecycle
    LR: 0.003
    WEIGHT_DECAY: 0.01
    MOMENTUM: 0.9

    MOMS: [0.95, 0.85]
    PCT_START: 0.4
    DIV_FACTOR: 10
    DECAY_STEP_LIST: [35, 45]
    LR_DECAY: 0.1
    LR_CLIP: 0.0000001

    LR_WARMUP: False
    WARMUP_EPOCH: 1

    GRAD_NORM_CLIP: 10