P2ANet
收藏资源简介:
P2ANet是一个大规模的乒乓球比赛视频动作检测基准数据集,由百度公司创建。该数据集包含2721个从世界乒乓球锦标赛和奥运会广播视频中收集的视频片段,总时长272小时。数据集通过与乒乓球专业人士和裁判合作,使用专门设计的标注工具箱进行精细动作标签(共14类)的标注。P2ANet数据集的特点是动作密集且快速移动,动作长度从0.3秒到3秒不等,超过90%的动作持续时间不到1秒,每10秒约有15个动作,密度极高。该数据集主要用于视频动作识别和定位研究,旨在解决快速移动和密集动作场景下的动作检测问题。
P2ANet is a large-scale benchmark dataset for table tennis match video action detection, created by Baidu. This dataset contains 2,721 video clips collected from the broadcast videos of World Table Tennis Championships and Olympic Games, with a total duration of 272 hours. The dataset was annotated with fine-grained action labels (14 categories in total) using a specially designed annotation toolbox, in collaboration with table tennis professionals and referees. Notably, P2ANet features dense and fast-paced actions, with action durations ranging from 0.3 seconds to 3 seconds. Over 90% of the actions last less than 1 second, and there are approximately 15 actions per 10 seconds, leading to an extremely high action density. This dataset is primarily used for research on video action recognition and localization, aiming to solve the action detection problems in fast-moving and dense action scenarios.

- 1P2ANet: A Dataset and Benchmark for Dense Action Detection from Table Tennis Match Broadcasting Videos百度公司 · 2024年



