MultiTHUMOS
收藏arXiv2025-09-30 收录
下载链接：
http://ai.stanford.edu/~syyeung/everymoment.html
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是对THUMOS'14动作检测数据集中的400个视频进行了密集的多标签帧级动作标注，总计30小时的视频时长，包含了38,690个标注，覆盖了65个动作类别。平均每帧有1.5个标签，每个视频有10.5个动作类别。该数据集的任务旨在进行在线动作检测和动作预期。
This dataset provides dense multi-label frame-level action annotations for 400 videos selected from the THUMOS'14 action detection dataset. It has a total cumulative video duration of 30 hours, contains 38,690 annotations, covers 65 action categories, and averages 1.5 labels per frame and 10.5 action categories per video. The supported tasks of this dataset are online action detection and action anticipation.
