tracks: AABB (M, 8), or OBB (M, 9) with angle after h

BoxMOT gives you one CLI and one Python API for running modern multi-object tracking workflows. It covers direct tracking, catalog-backed evaluation, tuning, research loops, ReID training and evaluation, and ReID export without forcing you to rebuild the detector and tracker stack for each experiment.

Why BoxMOT

• One interface for track, generate, eval, tune, research, train-reid, eval-reid, export, and native build workflows.
• Swappable trackers with shared detector and ReID plumbing.
• Dataset and experiment workflows with reusable detections and embeddings.
• Support for both AABB and OBB tracking paths.
• Optional production-ready native C++ tracker implementations with the same metrics as the Python path, opted into via --tracker-backend cpp and embeddable in standalone C++ projects via CMake (see Native C++ Integration).
• Public Python API for embedding the same workflows in applications and notebooks.

Installation

BoxMOT supports Python 3.10 through 3.13.

pip install boxmot
boxmot --help

The default package uses the standard PyPI PyTorch build. Source checkouts and CI can explicitly select the lockfile-backed cpu or cu130 profile. For those profiles and mode-specific extras such as yolo, service, evolve, research, onnx, openvino, and tflite, see the installation guide.

Docker images

Published images cover GPU and CPU CLI workflows plus separate CPU geometry and GPU ReID tracker services:

# GPU-enabled detector, CLI, evaluation, and interactive workflows
docker run --rm -it --gpus all boxmot/boxmot:latest

# The same CLI workflows on CPU
docker run --rm -it boxmot/boxmot:latest-cpu

# CPU-only stateful HTTP tracking from externally supplied detections
docker run --rm -p 8000:8000 boxmot/boxmot-service:latest

# CUDA/ReID stateful tracking from detections plus an encoded image per frame
docker run --rm --gpus all -p 8000:8000 \
  -v "$PWD/models/osnet_x0_25_msmt17.pt:/models/osnet_x0_25_msmt17.pt:ro" \
  -e BOXMOT_SERVICE_REID_WEIGHTS=/models/osnet_x0_25_msmt17.pt \
  boxmot/boxmot-service:latest-gpu

Versioned and commit-addressed tags are also published. GPU CLI tags are <version> and sha-<commit>; CPU CLI tags append -cpu. The CPU service uses canonical <version> and sha-<commit> tags in its own repository, while the GPU service appends -gpu. See the installation guide for local builds.

Both services accept ordered AABB or OBB detections and keep isolated state per stream/session; neither runs a detector. The CPU image supports ByteTrack, OCSort, and SFSORT without image pixels. The GPU image supports StrongSORT, BotSORT, DeepOCSORT, HybridSORT, BoostTrack, and OccluBoost, and requires a raw base64-encoded JPEG or PNG in image_base64 for every frame, including empty detection frames. See the deployment guide for the request schema and horizontal-scaling requirements.

Benchmark Results

Related guides:

• Evaluation and Postprocessing
• Experiment Workflows
• Native C++ Integration

Minimal Usage

CLI:

boxmot track --detector yolo26n --reid lmbn_n_duke --tracker occluboost --source 0 --save --show

Python:

import numpy as np
from boxmot.trackers import OccluBoost

tracker = OccluBoost()

# dets: (N, 6) array with [x1, y1, x2, y2, conf, cls] per detection
dets = np.array([[100, 200, 300, 400, 0.9, 0]], dtype=np.float32)
# OBB alternative: (N, 7) with [cx, cy, w, h, angle_radians, conf, cls]
# dets = np.array([[200, 300, 200, 200, 0.25, 0.9, 0]], dtype=np.float32)
img = np.zeros((480, 640, 3), dtype=np.uint8)  # current frame

# tracks: AABB (M, 8), or OBB (M, 9) with angle after h
tracks = tracker.update(dets, img)
print(tracks)

Contributing

Start with CONTRIBUTING.md and the contributor docs.

Contributors

Support and Citation

• Bugs and feature requests: GitHub Issues
• Questions and discussion: GitHub Discussions or Discord
• Limited free consulting is available for nonprofit nature conservation projects using BoxMOT. Contact box-mot@outlook.com to discuss your project.
• Citation metadata: CITATION.cff
• Commercial support: box-mot@outlook.com