boxmot

Why we included this project

If you build video analytics, or any system that has to keep the same identity attached to an object across frames, BoxMOT saves you from wiring up a tracker by hand. It puts several well-known trackers, including ByteTrack and BoT-SORT among others, behind one Python API and a CLI, so you can swap trackers without rewriting your detection or re-identification code. Because it handles both axis-aligned and oriented bounding boxes, it suits ordinary surveillance footage and rotated-object cases like drone imagery. The bundled workflows for benchmarking against standard MOT datasets, tuning, and re-identification training cover both research and production needs, and there's an optional C++ backend when you need lower latency. If you already use a YOLO-family detector, the drop-in integration makes adoption straightforward.