metafile.yml

Collections:
  - Name: OFA
    Metadata:
      Architecture:
        - ResNet
        - Transformer
      Training Data:
        - CC12M
        - CC3M
        - SBU
        - COCO
        - VG
        - VQAv2
        - GQA
        - RefCOCO
        - OpenImages
        - Object365
        - YFCC100M
        - ImageNet-21K
        - Pile
    Paper:
      Title: 'OFA: Unifying Architectures, Tasks, and Modalities Through a Simple
        Sequence-to-Sequence Learning Framework'
      URL: https://arxiv.org/abs/2202.03052
    README: configs/ofa/README.md

Models:
  - Name: ofa-base_3rdparty-finetuned_refcoco
    Metadata:
      FLOPs: null
      Parameters: 182238536
    In Collection: OFA
    Results:
      - Task: Visual Grounding
        Dataset: RefCOCO
        Metrics:
          Accuracy (testA): 90.49
          Accuracy (testB): 83.63
    Weights: https://download.openmmlab.com/mmclassification/v1/ofa/ofa-base_3rdparty_refcoco_20230418-2797d3ab.pth
    Config: configs/ofa/ofa-base_finetuned_refcoco.py
    Converted From:
      Weights: https://ofa-beijing.oss-cn-beijing.aliyuncs.com/checkpoints/refcoco_base_best.pt
      Code: https://github.com/OFA-Sys/OFA
  - Name: ofa-base_3rdparty-finetuned_vqa
    Metadata:
      FLOPs: null
      Parameters: 182238536
    In Collection: OFA
    Results:
      - Task: Visual Question Answering
        Dataset: VQAv2
        Metrics:
          Accuracy: 78.00   # Report from the official repo
    Weights: https://download.openmmlab.com/mmclassification/v1/ofa/ofa-base_3rdparty_coco-vqa_20230418-f38539a5.pth
    Config: configs/ofa/ofa-base_finetuned_vqa.py
    Converted From:
      Weights: https://ofa-beijing.oss-cn-beijing.aliyuncs.com/checkpoints/vqa_large_best.pt
      Code: https://github.com/OFA-Sys/OFA
  - Name: ofa-base_3rdparty-finetuned_caption
    Metadata:
      FLOPs: null
      Parameters: 182238536
    In Collection: OFA
    Results:
      - Task: Image Caption
        Dataset: COCO
        Metrics:
          BLEU-4: 42.64
          CIDER: 144.50
    Weights: https://download.openmmlab.com/mmclassification/v1/ofa/ofa-base_3rdparty_coco-caption_20230418-de18914e.pth
    Config: configs/ofa/ofa-base_finetuned_caption.py
    Converted From:
      Weights: https://ofa-beijing.oss-cn-beijing.aliyuncs.com/checkpoints/caption_base_best.pt
      Code: https://github.com/OFA-Sys/OFA
  - Name: ofa-base_3rdparty-zeroshot_vqa
    Metadata:
      FLOPs: null
      Parameters: 182238536
    In Collection: OFA
    Results:
      - Task: Visual Question Answering
        Dataset: VQAv2
        Metrics:
          Accuracy: 58.32
    Weights: https://download.openmmlab.com/mmclassification/v1/ofa/ofa-base_3rdparty_pretrain_20230418-dccfc07f.pth
    Config: configs/ofa/ofa-base_zeroshot_vqa.py
    Converted From:
      Weights: https://ofa-beijing.oss-cn-beijing.aliyuncs.com/checkpoints/ofa_base.pt
      Code: https://github.com/OFA-Sys/OFA