In Chapter 5, Neural Network Architecture and Models, we learned about DCGANs. They consist of a generator model and a discriminator model. The generator model takes in a random vector representing the feature of an image and runs through a CNN to produce an artificial image, G(z). Due to this, the generator model returns the absolute probability G(z), of generating a new image and its class. The discriminator (D) network is a binary classifier. It takes in the real image from a sample probability, distribution of images (p-data) and the artificial image from the generator in order to generate a probability, P(z), that the final image has been sampled from a real image distribution. Thus, the discriminator model returns the conditional probability that the class of the final image is from a given distribution.
The discriminator feeds the...
