Building realtime sports video analytics pipelines

Turning raw sports broadcast data into structured data is a problem touching almost every corner of applied computer vision: detection, tracking, segmentation, OCR and temporal reasoning, all under the real-time constraints of a live video stream. In this workshop we walk through a compact version of such pipeline. We cover automated data labeling and vision model fine-tuning, show how to turn raw model detections into higher-level information and match events, and how to evaluate the results against ground truth. Participants will leave with an understanding of how modern real-time vision pipelines are built from the ground up.