tjgreg92/esports-cv ? reverse-engineered prompt

Build me a Python project that watches Call of Duty League broadcast video and turns it into a live win probability overlay.

I want the app to pull useful game info from the video frame by frame, like the team scores, the timer, and player counts from the minimap, then use that data to predict which team is more likely to win. Use computer vision for the minimap detection, template matching for the scoreboard numbers, and a simple machine learning model for the win prediction.

Please include the full pipeline too, from preparing training images and labels, to training the detection model, extracting telemetry from match videos, validating the predictor, and rendering a demo overlay or short clip at the end. If you need to look up current docs online for anything, go ahead. I’d also like the code to work well on Apple Silicon if possible, and to include a clean way to run each step in order from raw video to final chart and presentation output.