SoccerNet
收藏资源简介:
SoccerNet是一个专注于足球视频中动作定位的大型可扩展数据集,由阿卜杜拉国王科技大学创建。该数据集包含500场完整的足球比赛视频,总时长764小时,涵盖2014至2017年六个主要欧洲联赛的三个赛季。数据集包含6,637个时间标注,主要分为进球、黄/红牌和换人三类事件。这些标注最初以一分钟分辨率自动从在线比赛报告中解析,随后手动细化至一秒分辨率。SoccerNet旨在解决长视频中非常稀疏事件的定位问题,并提供了用于检测足球事件的强大基准模型。数据集的应用领域包括足球视频分析、策略分析和球员评估等。
SoccerNet is a large-scale and scalable dataset dedicated to action localization in soccer videos, developed by King Abdullah University of Science and Technology (KAUST). This dataset includes 500 full-length soccer match videos with a total duration of 764 hours, covering three seasons of six major European leagues spanning from 2014 to 2017. It contains 6,637 temporal annotations, which are primarily categorized into three types of events: goals, yellow/red cards, and substitutions. These annotations were first automatically extracted from online match reports at a 1-minute resolution, and then manually refined to a 1-second resolution. SoccerNet aims to tackle the problem of localizing extremely sparse events in long videos, and provides robust baseline models for soccer event detection. Application areas of this dataset include soccer video analysis, tactical analysis, player evaluation, and other related fields.

- 1SoccerNet: A Scalable Dataset for Action Spotting in Soccer Videos阿卜杜拉国王科技大学 · 2018年



