VidHOI
收藏资源简介:
该数据集是迄今为止最大的视频数据集,其中包含了全面的人类-物体交互(HOI)注释。它汇集了来自社交媒体的、展现非剧本化人类活动在动态和非结构化环境中的视频。该数据集涵盖了78个物体类别和50种关系类别,其中包括动作和空间关系,其中一半涉及时间关系。其规模宏大,包含7122个视频、75.5万个HOI实例和730万关键帧。该数据集的任务是进行基于视频的人类-物体交互(V-HOI)检测。
This dataset is the largest video dataset to date, featuring comprehensive Human-Object Interaction (HOI) annotations. It collects videos from social media that showcase unscripted human activities in dynamic and unstructured environments. This dataset covers 78 object categories and 50 relationship categories, including both action and spatial relationships, half of which involve temporal relationships. With its massive scale, it contains 7,122 videos, 755,000 HOI instances, and 7.3 million keyframes. The core task of this dataset is video-based Human-Object Interaction (V-HOI) detection.




