Original work: https://www.pyimagesearch.com/2020/05/04/covid-19-face-mask-detector-with-opencv-keras-tensorflow-and-deep-learning/
I use MobileNetV2 pretrained model without trainable parameters. Then, I concatenate this model with serveral custom layers ( AveragePooling 2D, Flatten, Dense with 128 relu neurons an 0.5 dropout and last 2 dense relu neurons.