/
r50_lvis_paco_pascalpart_partimagenet.yaml
53 lines (53 loc) · 1.79 KB
/
r50_lvis_paco_pascalpart_partimagenet.yaml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
_BASE_: "../coco/r50_coco.yaml"
MODEL:
WEIGHTS: "models/r50_lvis_paco.pth"
ROI_HEADS:
NUM_CLASSES: 1203 # max of num_classes in all train datasets
ROI_BOX_HEAD:
MULT_OBJECT_SCORE: True
USE_SIGMOID_CE: True
USE_ZEROSHOT_CLS: True
USE_ZEROSHOT_CLS_GROUP: True
ZEROSHOT_WEIGHT_DIM: 1024
ZEROSHOT_WEIGHT_PATH_GROUP: [
'datasets/metadata/lvis_v1_clip_RN50_a+cname.npy',
'datasets/metadata/paco_clip_RN50_a+cname.npy',
'datasets/metadata/pascal_part_clip_RN50_a+cname.npy',
'datasets/metadata/partimagenet_clip_RN50_a+cname.npy'
]
ZEROSHOT_WEIGHT_INFERENCE_PATH: 'datasets/metadata/partimagenet_clip_RN50_a+cname.npy'
IGNORE_ZERO_CATS_GROUP: [False, False, False, False]
USE_FED_LOSS_GROUP: [True, True, False, False]
CAT_FREQ_PATH_GROUP: [
'datasets/metadata/lvis_v1_train_cat_info.json',
'datasets/metadata/paco_lvis_v1_train_cat_info.json',
'', ''
]
DATASETS:
TRAIN: ("lvis_v1_train", "paco_lvis_v1_train", "pascal_part_train", "partimagenet_train")
TEST: ("partimagenet_val",)
SOLVER:
IMS_PER_BATCH: 16
USE_CUSTOM_SOLVER: True
OPTIMIZER: "ADAMW"
BASE_LR: 0.00005
STEPS: (30000, 35000)
MAX_ITER: 40000
CHECKPOINT_PERIOD: 5000
INPUT:
CUSTOM_AUG: ResizeShortestEdge
MIN_SIZE_TRAIN_SAMPLING: range
MIN_SIZE_TRAIN: (640, 800)
DATALOADER:
SAMPLER_TRAIN: "MultiDatasetSampler"
DATASET_RATIO: [4, 2, 1, 1]
USE_DIFF_BS_SIZE: True
DATASET_BS: [2, 2, 2, 2]
USE_RFS: [True, True, False, False]
DATASET_MIN_SIZES: [[640, 800], [640, 800], [640, 800], [640, 800]]
DATASET_MAX_SIZES: [1333, 1333, 1333, 1333]
FILTER_EMPTY_ANNOTATIONS: False
MULTI_DATASET_GROUPING: True
DATASET_ANN: ['box', 'box', 'box', 'box']
NUM_WORKERS: 8
WITH_IMAGE_LABELS: True