Understanding broadcast videos is a challenging task in computer vision, as\nit requires generic reasoning capabilities to appreciate the content offered by\nthe video editing. In this work, we propose SoccerNet-v2, a novel large-scale\ncorpus of manual annotations for the SoccerNet video dataset, along with open\nchallenges to encourage more research in soccer understanding and broadcast\nproduction. Specifically, we release around 300k annotations within SoccerNet's\n500 untrimmed broadcast soccer videos. We extend current tasks in the realm of\nsoccer to include action spotting, camera shot segmentation with boundary\ndetection, and we define a novel replay grounding task. For each task, we\nprovide and discuss benchmark results, reproducible with our open-source\nadapted implementations of the most relevant works in the field. SoccerNet-v2\nis presented to the broader research community to help push computer vision\ncloser to automatic solutions for more general video understanding and\nproduction purposes.\n