The new Microsoft neural network is called CariGAN. It makes from photos and videos of people terrifying images that are difficult to distinguish from hand-drawn ones.
The neural network consists of two parts: CariGeoGAN is responsible for determining the shape of the face and creating a caricature model. And the second part of the program CariStyGan is used to create the artistic style of photos. Developers had to train a neural network using thousands of hand-drawn caricatures.
The effectiveness of the finished program was checked by two parameters, according to The Next Web. The neural network, like a good caricature, had to preserve the details of the original photo so that it was possible to determine a person’s personality. They also checked how the finished result looked like real caricatures, many of the survey could not distinguish them from the hand-drawn ones.