🧑‍💻 Face Tracking with MTCNN + BoxMot

This project performs multi-face tracking using MTCNN for face detection and BoxMot for tracking.

---

📽️ Demo

• Test Video: MOT16-08 from the MOTChallenge benchmark
• Trackers tested: BoT-SORT, ByteTrack, OCSORT (via BoxMot)

🛠️ Features

• ✅ Face detection with MTCNN
• ✅ Tracker selection: BoT-SORT / ByteTrack / OCSORT
• ✅ Confidence-based face filtering
• ✅ Real-time thumbnail gallery for tracked faces
• ✅ Modular Python codebase

🗂️ Project Structure

face_tracking_project/
├── main.py                    # Entry point
├── arguments.py               # CLI options
├── face_tracking/
│   ├── tracking_runner.py     # Main tracking logic
│   ├── select_tracker.py      # Tracker selection
│   └── gallery_drawer.py      # Draw face thumbnails

🚀 How to Run

1. Install dependencies

pip install facenet-pytorch boxmot opencv-python

2. Run the tracker

python main.py \
  --video_path ./assets/source_lab.mp4 \
  --tracker botsort \
  --reid_weights mobilenetv2_x1_4_market1501.pt \
  --output_path output_custom_draw.mp4

⚙️ CLI Options

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

🎬 Tracking Demo

🔹 [Original Video (MOT16-08)]

MOT16-08-raw.webm

🔸 [Tracking Result (BoT-SORT)]

output_custom_draw.webm

📌 References

• Dataset: MOTChallenge - MOT16-08
• Tracker Library: boxmot (GitHub)