MetaVDでは、UCF101, HMDB51, ActivityNet, STAIR Actions, Charades, Kinetics-700が取り込まれています。
MetaVD is a meta-dataset for utilizing various action recognition datasets by connecting between them.
In particular, MetaVD provides human-annotated relations, such as equality, similarity, and hierarchy, between action labels across the action recognition datasets based on the meaning of the action labels which each dataset defines.
MetaVD v1 contains UCF101, HMDB51, ActivityNet, STAIR Actions, Charades, and Kinetics-700. It will be expanded.



  • Yuya Yoshikawa, Yutaro Shigeto, and Akikazu Takeuchi, “MetaVD: A Meta Video Dataset for enhancing human action recognition datasets,” Computer Vision and Image Understanding, vol. 212, p. 103276, Nov. 2021.

    MetaVDの提案と、MetaVDを用いて学習した動作認識モデルの性能評価を行いました。 We proposed MetaVD, and evaluated the accuracy of the action recognition models trained on MetaVD.


動作ラベル間の関係性ネットワークを見ることができます。 動作ラベルを検索でき、その動作ラベルと関連がある他の動作ラベルも合わせて見つけることができます。 You can see the relation network defined on MetaVD, and search action labels contained on MetaVD.


( ※動画データは別途ダウンロードが必要です。 )
You can download the MetaVD dataset that contains human-annotated relations between action labels across the existing action recognition datasets.
Note: since it does NOT contain the video data of each dataset in MetaVD, please donwload the video data from the repository of each dataset if necessary.