Pluggable Python and C++ multi-object tracking modules for axis-aligned and oriented bounding box detections from any model.
Docs • Installation • Modes • API Reference • Trackers • Contributing
BoxMOT provides independent detector, segmentor, appearance-encoder, and tracker components built around validated Torch structures. Pipelines compose those components; the CLI owns sources, outputs, materialized datasets, evaluation, tuning, research, and ReID workflows.
- One interface for
track,materialize,materialize --time-variant,eval,tune,research,train-reid,eval-reid,compare-reid,export,export-edgetam,install, and nativebuildworkflows. - Swappable components with explicit capabilities and requirements.
- Immutable, keyed Parquet builds with reusable detections, masks, 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 cppand embeddable in standalone C++ projects via CMake (see Native C++ Integration and platform dependency installation). - A structured Python API for embedding components and pipelines in applications.
BoxMOT supports Python 3.10 through 3.13.
pip install boxmot
boxmot --helpThe 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.
| Tracker key | Status | MOT17 ablation | SportsMOT val | MMOT OBB test | OBB | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| HOTA | MOTA | IDF1 | HOTA | MOTA | IDF1 | HOTA | MOTA | IDF1 | |||
| occluboost | ✅ | 71.10 (71.10) |
78.50 (78.50) |
85.28 (85.28) |
83.17 | 97.48 | 89.36 | 49.84 (49.84) |
39.41 (39.41) |
58.60 (58.60) |
✅ |
| botsort | ✅ | 69.68 (69.74) |
78.23 (78.27) |
82.33 (82.55) |
76.93 | 98.11 | 78.30 | 52.31 (52.40) |
45.43 (45.53) |
61.42 (61.42) |
✅ |
| boosttrack | ✅ | 69.25 (—) |
75.91 (—) |
83.20 (—) |
76.32 | 97.08 | 77.82 | 48.39 (—) |
41.36 (—) |
56.36 (—) |
✅ |
| strongsort | ✅ | 68.05 (—) |
76.19 (—) |
80.76 (—) |
79.80 | 97.31 | 80.27 | 49.76 (—) |
43.70 (—) |
57.32 (—) |
✅ |
| deepocsort | ✅ | 67.95 (—) |
75.83 (—) |
80.54 (—) |
79.51 | 97.94 | 79.59 | 50.84 (—) |
44.21 (—) |
59.33 (—) |
✅ |
| bytetrack | ✅ | 67.68 (67.68) |
78.04 (78.04) |
79.16 (79.16) |
67.93 | 97.25 | 76.90 | 33.97 (33.97) |
33.72 (33.72) |
39.74 (39.74) |
✅ |
| hybridsort | ✅ | 67.31 (—) |
74.09 (—) |
78.87 (—) |
81.14 | 98.07 | 81.88 | 54.64 (—) |
47.50 (—) |
64.67 (—) |
✅ |
| ocsort | ✅ | 66.44 (66.44) |
74.55 (74.55) |
77.90 (77.90) |
76.34 | 96.60 | 75.64 | 28.64 (28.64) |
26.17 (26.17) |
30.06 (30.06) |
✅ |
| sfsort | ✅ | 62.65 (62.65) |
76.87 (76.87) |
69.18 (69.18) |
75.73 | 98.39 | 72.99 | 47.83 (47.83) |
45.42 (45.42) |
52.09 (52.09) |
✅ |
Scores are Python first and C++ in parentheses.
MMOT reported metrics are 'class average'. See Experiment Workflows for details.
Related guides:
CLI:
boxmot track --detector yolo26n --reid lmbn_n_duke --tracker occluboost \
--source 0 --save --showEvaluate a tracker:
boxmot eval \
--dataset mot17 \
--split ablation \
--detector yolox-x-mot17 \
--reid lmbn-n-duke \
--tracker botsortPython with axis-aligned boxes (AABB):
import numpy as np
from boxmot import OccluBoost
tracker = OccluBoost()
dets = np.array([[100, 200, 300, 400, 0.9, 0]])
frame = np.zeros((480, 640, 3), dtype=np.uint8) # BGR image
tracks = tracker.update(dets, frame)
print(tracks[:, 4].astype(int)) # track IDsPython with oriented boxes (OBB):
import numpy as np
from boxmot import OccluBoost
tracker = OccluBoost(is_obb=True)
# cx, cy, width, height, angle in radians, confidence, class_id
dets = np.array([[200, 300, 200, 100, np.pi / 6, 0.9, 0]])
frame = np.zeros((480, 640, 3), dtype=np.uint8) # BGR image
tracks = tracker.update(dets, frame)
print(tracks[:, 5].astype(int)) # track IDs2D/3D sensor fusion with ego-motion compensation using EagerMot:
import numpy as np
from boxmot import EagerMot
tracker = EagerMot()
camera = np.array([ # synthetic camera projection
[100, 0, 100, 0],
[0, 100, 50, 0],
[0, 0, 1, 0],
])
camera_to_world = np.eye(4)
camera_to_world[0, 3] = 1.0 # synthetic pose: camera at world x = 1 meter
dets = np.array([[78, 39, 122, 62, 0.95, 0]])
# 3D in camera coordinates: x, y, z, yaw, length, width, height, confidence, class_id
dets_3d = np.array([[0, 1, 10, 0, 4, 2, 2, 0.9, 0]])
tracks_2d, tracks_3d = tracker.update(
dets,
detections_3d=dets_3d,
camera=camera,
image_size=(100, 200), # height, width
camera_to_world=camera_to_world,
)
print(tracks_2d[:, 4].astype(int)) # image track IDs
print(tracks_3d[:, 7].astype(int)) # 3D track IDsReplace the synthetic pose with the current frame's absolute 4×4 camera-to-world
transform on every update. KITTI evaluation loads these from poses.npy when
the dataset declares a poses modality. If poses are unavailable, omit
camera_to_world throughout the sequence to track in camera coordinates.
Start with CONTRIBUTING.md and the contributor docs.
- 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
[email protected]to discuss your project. - Citation metadata: CITATION.cff
- Commercial support:
[email protected]
