MetaVD: A Meta Video Dataset for enhancing human action recognition datasets

作者：

Highlights：

• We construct a meta video dataset for human action recognition, called MetaVD.

• MetaVD integrates UCF101, HMDB51, ActivityNet, STAIR Actions, Charades and Kinetics-700.

• MetaVD provides four types of relations between action labels across datasets.

• We present simple methods to enhance datasets using MetaVD.

• Models trained on datasets enhanced by MetaVD achieve high recognition performances.

摘要

•We construct a meta video dataset for human action recognition, called MetaVD.•MetaVD integrates UCF101, HMDB51, ActivityNet, STAIR Actions, Charades and Kinetics-700.•MetaVD provides four types of relations between action labels across datasets.•We present simple methods to enhance datasets using MetaVD.•Models trained on datasets enhanced by MetaVD achieve high recognition performances.

论文关键词：Human action recognition,Video datasets

论文评审过程：Received 16 November 2020, Revised 7 July 2021, Accepted 4 September 2021, Available online 13 September 2021, Version of Record 23 September 2021.

论文官网地址：https://doi.org/10.1016/j.cviu.2021.103276