“We are using just one model now for the three types of faces ... tasks and train a convolutional neural network (CNN) using PyTorch “to analyze each frame and produce a pixel-accurate map ...