roboflow/sports — reverse-engineered prompt

Build me a simple Python sports vision toolkit that works from video and focuses on sports analytics, especially soccer. I want reusable tools that can detect players, detect the ball, find pitch or court keypoints, and make it possible to track players and the ball over time. If it makes sense, also support reading jersey numbers from basketball footage, since that seems to be part of the project too.

Please make it feel like a real usable repo, not just a one off experiment. Include a few example workflows that take a video, run the models, and output visual overlays or useful analytics style views. It should be easy to use datasets and models from Roboflow Universe, and easy to extend later for harder problems like player re identification or camera calibration.

Keep it installable from source in a Python 3.8 or newer environment, and add clear setup and usage docs. If you need details, check the current Roboflow docs and use the examples in the repo as a guide.