SoccerNet: building the largest and most comprehensive open-source dataset and benchmark for sports video analysis
Anthony Cioppa, Assistant Professor @ University of Liège
Artificial intelligence is transforming sports analysis. Since 2018, SoccerNet has established itself as the largest and most comprehensive open dataset, resulting from a collaboration initiated by KAUST (Saudi Arabia) and the University of Liège (Belgium), and now strengthened by many academic partners (Aalborg University, UCLouvain, EPFL, etc.) and industrial partners (EVS, Sportradar, Spiideo, Footovision, etc.). With more than 850 hours of manually annotated videos and over 15 million annotations (actions, players, cameras), SoccerNet provides a unique resource for developing and comparing computer vision methods, covering tasks such as action spotting, camera calibration, and player tracking. Beyond the data itself, SoccerNet fosters a dynamic international community: more than 1,800 researchers on its Discord server, annual challenges that have attracted over 1,500 submissions from 300 teams, and more than 30 publications. The success of SoccerNet illustrates the importance of open data and shared benchmarks in driving scientific innovation. The talk will finally focus on the concrete applications that directly emerged from these innovations, from enhanced broadcasting for the viewers to improved well-being of the athletes.