This project performs multi-face tracking using MTCNN for face detection and BoxMot for tracking.
- Test Video: MOT16-08 from the MOTChallenge benchmark
- Trackers tested:
BoT-SORT,ByteTrack,OCSORT(via BoxMot)
- β Face detection with MTCNN
- β Tracker selection: BoT-SORT / ByteTrack / OCSORT
- β Confidence-based face filtering
- β Real-time thumbnail gallery for tracked faces
- β Modular Python codebase
face_tracking_project/
βββ main.py # Entry point
βββ arguments.py # CLI options
βββ face_tracking/
β βββ tracking_runner.py # Main tracking logic
β βββ select_tracker.py # Tracker selection
β βββ gallery_drawer.py # Draw face thumbnails
- Install dependencies
- Run the tracker
pip install facenet-pytorch boxmot opencv-pythonpython main.py \
--video_path ./assets/source_lab.mp4 \
--tracker botsort \
--reid_weights mobilenetv2_x1_4_market1501.pt \
--output_path output_custom_draw.mp4
| Argument | Description | Default |
|---|---|---|
--video_path | Path to input video | ./data/MOT16-08-raw.webm |
--output_path | Path to save result video | output_custom_draw.mp4 |
--tracker | Tracking model (botsort / bytetrack / ocsort) | botsort |
--reid_weights | Path to re-ID model (BoT-SORT only) | mobilenetv2_x1_4_market1501.pt |
--conf_thresh | Minimum confidence for detection | 0.7 |
--max_faces | Max face thumbnails in gallery | 6 |
πΉ [Original Video (MOT16-08)]
MOT16-08-raw.webm
πΈ [Tracking Result (BoT-SORT)]
output_custom_draw.webm
- Dataset: MOTChallenge - MOT16-08
- Tracker Library: boxmot (GitHub)