SoccerNet-v2
收藏资源简介:
SoccerNet-v2是由列日大学等多个研究机构合作创建的大型足球视频数据集,包含500个未剪辑的广播足球视频,总计764小时的视频内容。数据集通过手动标注了约30万个时间戳,涵盖了17种足球动作、13种摄像机镜头类型以及重播定位等任务。创建过程中,33名经常观看足球的标注者参与了约1600小时的标注工作,确保了标注的质量和一致性。SoccerNet-v2旨在推动计算机视觉技术在自动视频理解和编辑领域的应用,特别是在足球广播视频的理解和分析上,为自动化视频编辑和内容分析提供了丰富的数据资源。
SoccerNet-v2 is a large-scale soccer video dataset co-created by the University of Liège and multiple other research institutions. It contains 500 unedited broadcast soccer videos, totaling 764 hours of video content. The dataset features approximately 300,000 manually annotated timestamps, covering 17 categories of soccer actions, 13 types of camera shots, and tasks including replay localization. During the dataset's development, 33 annotators who regularly watch soccer matches dedicated approximately 1,600 hours to the annotation process, ensuring the quality and consistency of the annotated data. SoccerNet-v2 is designed to advance the application of computer vision technologies in the field of automatic video understanding and editing, particularly for the comprehension and analysis of soccer broadcast videos, and provides a rich data resource for automated video editing and content analysis.
- 1SoccerNet-v2: A Dataset and Benchmarks for Holistic Understanding of Broadcast Soccer Videos列日大学 · 2021年



