SportsHHI

SportsHHI数据集专注于体育视频中的人与人交互检测，由国家软件新技术重点实验室创建。该数据集包含34种高级交互类别，主要来自篮球和排球运动，共标注了11,398个关键帧中的118,075个人体边界框和50,649个交互实例。SportsHHI旨在通过详细的时空上下文推理，理解复杂的多人视频中的高级交互，如战术协作或对抗，推动视频视觉关系检测中时空上下文建模技术的发展。

The SportsHHI dataset focuses on human-human interaction detection in sports videos, and was developed by the State Key Laboratory of Novel Software Technology. It encompasses 34 high-level interaction categories, primarily sourced from basketball and volleyball, with a total of 118,075 human bounding boxes and 50,649 interaction instances annotated across 11,398 key frames. The core objective of SportsHHI is to comprehend high-level interactions such as tactical collaboration or confrontation in complex multi-person videos via detailed spatio-temporal contextual reasoning, thereby advancing the development of spatio-temporal context modeling technologies for video visual relationship detection.

搜集汇总

背景与挑战

背景概述

SportsHHI是一个用于体育视频中的人-人交互检测的数据集，源自CVPR 2024论文。该数据集旨在支持计算机视觉领域的研究，特别是针对体育场景下的交互识别任务，并提供了基线方法和评估代码。