_BASE_: "Base-C2_L_R5021k_640b64_4x.yaml"
MODEL:
  ROI_BOX_HEAD:
    USE_ZEROSHOT_CLS: True
    IMAGE_LABEL_LOSS: 'max_size'
  BACKBONE:
    NAME: build_swintransformer_fpn_backbone
  SWIN:
    SIZE: B-22k
  FPN:
    IN_FEATURES: ["swin1", "swin2", "swin3"]
  WEIGHTS: "models/BoxSup-C2_Lbase_CLIP_SwinB_896b32_4x.pth"
SOLVER:
  MAX_ITER: 180000
  IMS_PER_BATCH: 32
  BASE_LR: 0.0001
  WARMUP_ITERS: 1000
  WARMUP_FACTOR: 0.001
DATASETS:
  TRAIN: ("lvis_v1_train_norare","imagenet_lvis_v1")
DATALOADER:
  SAMPLER_TRAIN: "MultiDatasetSampler"
  DATASET_RATIO: [1, 4]
  USE_DIFF_BS_SIZE: True
  DATASET_BS: [4, 16]
  DATASET_INPUT_SIZE: [896, 448]
  USE_RFS: [True, False]
  DATASET_INPUT_SCALE: [[0.1, 2.0], [0.5, 1.5]]
  FILTER_EMPTY_ANNOTATIONS: False
  MULTI_DATASET_GROUPING: True
  DATASET_ANN: ['box', 'image']
  NUM_WORKERS: 8
WITH_IMAGE_LABELS: True