Face Anonymizer

Video face anonymization using MediaPipe and YOLOv8-Face.

Features

• ✅ Real-time face detection with MediaPipe
• ✅ 2 detectors: MediaPipe (selfies) & YOLOv8-Face (distant faces)
• ✅ 3 anonymization methods: Blur, Pixelate, Black
• ✅ Works with webcam or video files
• ✅ Automatic output naming with timestamps
• ✅ Configurable padding and confidence

Installation

# Clone the repo
git clone https://github.com/Moncef-Bj/face-anonymizer.git
cd face-anonymizer

# Create virtual environment
uv venv --python 3.11
.venv\Scripts\activate  # Windows
source .venv/bin/activate  # Linux/Mac

# Install dependencies
uv pip install -e .

Usage

python src/face_anonymizer/cli.py INPUT [method] [padding] [detector]

Arguments

Argument	Options	Default	Description
INPUT	file.mp4 or 0	required	Video file or webcam
method	blur, pixelate, black	blur	Anonymization method
padding	0.0 - 1.0	0.3	Margin around faces
detector	mediapipe, yolo	mediapipe	Detection model

Detectors

Detector	Best for	Speed
mediapipe	Selfies, webcam, close faces	Fast
yolo	Videos, distant faces, crowds	Accurate

Examples

# Webcam with MediaPipe (default, good for selfies)
python src/face_anonymizer/cli.py 0

# Webcam with YOLOv8-Face (better for distant faces)
python src/face_anonymizer/cli.py 0 blur 0.3 yolo

# Video file with pixelate effect
python src/face_anonymizer/cli.py video.mp4 pixelate 0.4 yolo

# Video with black rectangles
python src/face_anonymizer/cli.py input.mp4 black 0.3 mediapipe

Project Structure

face-anonymizer/
├── src/face_anonymizer/
│   ├── __init__.py      # Package initialization
│   ├── detector.py      # MediaPipe & YOLOv8 detectors
│   ├── anonymizer.py    # Blur, pixelate, black methods
│   ├── pipeline.py      # Video processing pipeline
│   └── cli.py           # Command-line interface
├── models/
│   └── yolov8n-face.pt  # YOLOv8-Face model
├── pyproject.toml       # Package configuration
├── requirements.txt     # Dependencies
└── README.md

Limitations

• MediaPipe works best with close-up frontal faces
• YOLOv8-Face better for varied conditions but slower
• Very small faces may still be missed

Future Improvements

• Add face tracking for consistent IDs across frames
• Add GUI interface (Gradio/Streamlit)
• Support batch processing
• Add Docker support

License

MIT