mmcm/vision/feature_extractor/wenlan/cfg/test_xyb.yml

MODEL:
    NAME: VL

    ENCODER: 'hfl/chinese-roberta-wwm-ext'
    TEXT_FEATURE_DIM: 768
    TEXT_TRANSFORMER_HEAD: 4
    TEXT_TRANSFORMER_LAYER: 4
    MAX_TEXT_LEN: 80
    
    CNN: 'tf_efficientnet_b5_ns'
    IMG_SIZE: 456
    IMG_FEATURE_DIM: 2048
    IMG_TRANSFORMER_HEAD: 4
    IMG_TRANSFORMER_LAYER: 4
    MAX_IMG_LEN: 37
    BOX_GRID: 6

    TOPK: 1
    QUEUE_SIZE: 9600
    MOMENTUM: 0.99
    TEMPERATURE: 0.07
    GRID_SIZE: 4
    IS_EXTRACT: True

DATASET:
    NAME: 'XYBDataset_all' #['AICDataset', 'VLDataset', 'superDataset', 'AICDataset_boxes']
    JSONPATH: '../data/jsonls/example.jsonl'
    WORKERS: 8