Lightweight face-mask detection using transfer learning (MobileNetV2 / VGG16 examples).
Fast inference with OpenCV DNN for face detection + a small TensorFlow classifier for mask vs no-mask.
During COVID, our campus had a camera system that flagged mask violations, but it was slow and unreliable.
This project is a quick experiment to build a fast, deployable detector: using OpenCV’s DNN for face localization and a lightweight TensorFlow MobileNetV2 model for mask detection.
The goal: portable, low-latency inference and reproducible results even on modest hardware.
Recommended: use Python 3.10 (TensorFlow 2.15 + NumPy 1.26 compatibility)
# create and activate venv
/Users/you/anaconda3/bin/python3.10 -m venv ./facemask
source ./facemask/bin/activate
# upgrade pip build tools
pip install --upgrade pip setuptools wheel
# install dependencies
pip install -r requirements.txt
# Run the detector:
python src/detect_mask_video.pyTo extend this repo:
- Add new model checkpoints in
models/ - Create a Flask API under
server/ - Dockerize with
Dockerfile+docker-compose.yml
Pull requests are welcome.
MIT License (or your preferred license).