Roboflow Sports is an open-source collection of reusable computer-vision tools for sports analytics, built by Roboflow as a testing ground for pushing object detection, image segmentation, keypoint detection and foundation models to their limits. It tackles the hard problems of sports footage — tracking a small, fast ball, detecting players and keypoints, and turning matches into data — with code designed to be reused in sports and beyond. The repository has about 5,400 stars on GitHub and is actively maintained, with updates in 2026.
In sports, every centimeter and every second matter, and the vision problems are genuinely hard: balls are tiny and fast, players occlude each other constantly, and useful analytics require consistent identities across an entire match. This repository packages Roboflow’s solutions and experiments in those areas — ball tracking, player detection and tracking, keypoint detection and image embeddings among them — as reusable components rather than one-off demos. It sits within Roboflow’s wider open-source ecosystem alongside their notebooks, inference server and related projects, so components here can feed into full production pipelines. Whether you are analyzing football (soccer) matches or adapting the trackers to another sport, the repo provides working code and a community of contributors attacking the same challenges.
Roboflow Sports is free and open source under the MIT license. The code in this repository costs nothing to use; Roboflow’s separate hosted platform has its own plans, but it is not required to use these tools.
Roboflow Sports is for sports-analytics developers, data teams at clubs and media companies, and computer-vision engineers who want proven components for tracking players and balls instead of starting from a blank repository.