DeepSportradar-v1: Computer Vision Dataset for Sports Understanding with High Quality Annotations

With the recent development of Deep Learning applied to Computer Vision,\nsport video understanding has gained a lot of attention, providing much richer\ninformation for both sport consumers and leagues. This paper introduces\nDeepSportradar-v1, a suite of computer vision tasks, datasets and benchmarks\nfor automated sport understanding. The main purpose of this framework is to\nclose the gap between academic research and real world settings. To this end,\nthe datasets provide high-resolution raw images, camera parameters and high\nquality annotations. DeepSportradar currently supports four challenging tasks\nrelated to basketball: ball 3D localization, camera calibration, player\ninstance segmentation and player re-identification. For each of the four tasks,\na detailed description of the dataset, objective, performance metrics, and the\nproposed baseline method are provided. To encourage further research on\nadvanced methods for sport understanding, a competition is organized as part of\nthe MMSports workshop from the ACM Multimedia 2022 conference, where\nparticipants have to develop state-of-the-art methods to solve the above tasks.\nThe four datasets, development kits and baselines are publicly available.\n