codedao360 / humanface-mask-detector

Build, train and deploy a complete human face mask detector using tensorflow.keras and opencv

How does human face mask detector work?

• 

• Result of human face detector

• reference for images: First Pictures, second Picture and Third Picture

Install dependencies

• tensorflow
• numpy
• sklearn (scikit-learn)
• tensorflow-gpu
• matplotlib
• imutils

Clone the project

$git clone https://github.com/bm777/humanface-mask-detector.git
cd humanface-mask-detector

How to implement your own face mask detector

• Fine tuning Load model: res10_300x300_ssd_iter_140000.caffemodel. We Construct new FC head(Fully connected layers), append it the base model and we freeze the base layers of the network(res10.*.caffemodel) -Notice: the weight sof the layers(base model) will not uploaded during the process of backpropagation. Whereas the head layer weight will be tunued.
• Training, testing and validation stage In this section, we load face mask dataset from disk and then train a model using tensorflow.keras 2.1.0. To perform your accuracy, you can change your hyperparameters(batch size and number of epoch learning rate)

python3 fm_train.py --dataset dataset
# After training you can test your classifier by:
python3 detect_mask_picture.py --image examples/other.jpg

• History ploted

Credit

• the dataset was created by : Prajna Bhandary
• part of code from Adrian Rosebrock
• part of fine tuning from this : LINK
• part of face Landmark to augment database: ALready done by Adrian Rosebrock
• pretrained model used: MobileNetV2 -> ImageNet weights