VLDet_OVCOCO_CLIP_R50_1x_caption.yaml

_BASE_: "Base_OVCOCO_C4_1x.yaml"
MODEL:
  SHARE_PROJ_V_DIM: 2048
  WEIGHTS: "models/coco_base.pth"
  WITH_CAPTION: True
  SYNC_CAPTION_BATCH: False
  SHARE_PROJ_L_DIM: 1024
  ROI_HEADS:
    NUM_CLASSES: 65 
  ROI_BOX_HEAD:
    WS_NUM_PROPS: 32
    ADD_IMAGE_BOX: True
    NEG_CAP_WEIGHT: 1.0
    OT_LOSS_WEIGHT: 0.01
    USE_CAPTION: True
    USE_OT: 'contrastive'
    ZEROSHOT_WEIGHT_PATH: 'datasets/coco/VLDet/coco_nouns_4764_emb.pth' 
    DETECTION_WEIGHT_PATH: 'datasets/coco/VLDet/coco_65_cls_emb.pth'
    CAT_FREQ_PATH: 'datasets/coco/zero-shot/instances_train2017_seen_2_del_cat_info.json' 
    ZEROSHOT_WEIGHT_DIM: 1024
    CAPTION_WEIGHT: 1.0
SOLVER:
  IMS_PER_BATCH: 32
  BASE_LR: 0.02
  STEPS: (60000, 80000)
  CHECKPOINT_PERIOD: 10000
  MAX_ITER: 90000
  CLIP_GRADIENTS:
    ENABLED: True
DATASETS:
  TRAIN: ("coco_zeroshot_train_del", "coco_caption_nouns_train_4764tags",) 
  TEST: ("coco_generalized_del_val",)
INPUT:
  CUSTOM_AUG: ResizeShortestEdge
  MIN_SIZE_TRAIN_SAMPLING: range
  MIN_SIZE_TRAIN: (800, 800)
DATALOADER:
  SAMPLER_TRAIN: "MultiDatasetSampler"
  DATASET_RATIO: [1, 4]
  USE_DIFF_BS_SIZE: True
  DATASET_BS: [2, 8]
  USE_RFS: [False, False]
  DATASET_MIN_SIZES: [[800, 800], [400, 400]]
  DATASET_MAX_SIZES: [1333, 667]
  FILTER_EMPTY_ANNOTATIONS: False
  MULTI_DATASET_GROUPING: True
  DATASET_ANN: ['box', 'caption']
  NUM_WORKERS: 8
WITH_IMAGE_LABELS: True
OUTPUT_DIR: output/test
FP16: True
TEST:
  EVAL_PERIOD: 10000