Charades

该数据集包含了9,848个日常室内活动视频，涉及与46类物体的互动，并包含一个由30个动词组成的词汇表，进而形成了157个动作类别。视频的平均时长为30秒，涵盖了15种室内场景类型。任务包括在线动作检测和动作预期。

This dataset comprises 9,848 videos of daily indoor human activities, involving interactions with 46 categories of objects. It includes a vocabulary of 30 verbs, which yields 157 unique action categories. The average duration of each video is 30 seconds, and the dataset covers 15 types of indoor scene contexts. The supported downstream tasks include online action detection and action anticipation.