I have hour-long MP4 recordings of complete basketball games and need a repeatable script that pinpoints every made basket in each file. The core requirement is simple: scan the full-game video, detect when the ball goes through the hoop, and return usable information (at minimum the exact timestamps; optional but welcome are frame numbers or a lightweight event log I can sync with other data). Key parameters • Input: single-camera, full-court MP4 footage recorded at standard broadcast resolution and frame rate. • Output: a list or JSON/CSV file of “made basket” events, one line per score. • Accuracy: false positives should be minimal; I can tolerate an occasional miss but would prefer recall above 90 %. I am open to your preferred tool chain—OpenCV, PyTorch, TensorFlow, or other CV/AI libraries—provided the solution runs on a modern Windows or Linux workstation without exotic hardware dependencies. Please build the model or rule-set into a self-contained Python script (command-line driven is fine) and document any weights, config files, or environment steps needed to reproduce results. To help you start, I can share several representative MP4 clips plus my manual annotation of scoring moments for validation. I will also test the finished script on unseen games and will consider the job complete once it flags made baskets within the specified accuracy and processes a 40-minute game in reasonable time. If you have prior work in sports video analytics or object-tracking models, let me know; code samples or brief demos will speed up selection.