Detecting face mask with OpenCV and TensorFlow. Using simple CNN or model provided by TensorFlow as MobileNetV2, VGG16, Xception.

Raw data collected from kaggle and script crawl_image.py, split to 'Mask' and 'Non Mask' class.

Using build_data.py to extract faces from raw dataset and resize to 64x64.

Clone the repo

git clone [email protected]:ksvbka/face-mask-detector.git

cd to project folder and create virtual env

virtualenv .env
source .env/bin/activate
pip install -r requirements.txt

Download raw dataset and execute script build_dataset.py to preprare dataset for training

cd data
bash download_data.sh
cd -
python3 build_dataset.py --data-dir data/dataset_raw/ --output-dir data/64x64_dataset

Execute train.py script and pass network architecture type dataset dir and epochs to it. Default network type is MobileNetV2.

python3 train.py --net-type MobileNetV2 --data-dir data/64x64_dataset --epochs 20

View tensorboard

tensorboard --logdir logs --bind_all

python3 mask_detect_image.py -m results/MobileNetV2-size-64-bs-32-lr-0.0001.h5 -i demo_image/2.jpg

Hyperparameter: - batch size: 32 - Learing rate: 0.0001 - Input size: 64x64x3

Model result

Download pre-trained model: link

Using MobileNetV2 model