NirAharon/BoT-SORT

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1522 stars · 494 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

BoT-SORT is a state-of-the-art multi-object tracker that combines motion and appearance information with camera motion compensation and an improved Kalman filter state vector. It supports YOLOX and YOLOv7 detectors, multi-class tracking, and an optional Re-identification variant, ranking first on MOT17 and MOT20 benchmarks.

Use cases

• track pedestrians in video with unique IDs
• multi-object tracking on MOT17/MOT20 benchmarks
• track multiple object classes with YOLOX or YOLOv7 detections
• compensate camera motion when tracking moving scenes
• re-identify people across occlusions in surveillance footage
• compare tracker performance using MOTA, IDF1, and HOTA metrics

When to choose

• you need top-accuracy pedestrian or multi-class tracking with a PyTorch pipeline
• your videos have camera motion requiring compensation
• you want appearance-based re-identification to keep IDs stable through occlusions
• you are benchmarking trackers on MOTChallenge datasets

When to avoid

• you need production deployment code, which is not yet provided
• you track objects without a compatible YOLO detector or want a detector-agnostic tracker
• you need a lightweight real-time tracker on CPU-only hardware
• you require Windows support out of the box (tested on Ubuntu)

Facets

library · maturity active

computer-vision machine-learning deep-learning computer-vision machine-learning deep-learning python multi-object-tracking tracking-by-detection pedestrian-tracking kalman-filter camera-motion-compensation re-identification yolox yolov7 pytorch motchallenge linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="NirAharon/BoT-SORT")