>>10846760This is a GAN (Generative Adversarial Network) architecture.
GAN consist of a Genarator and a Discriminator.
The Genarator is generating fake images.
The Discriminator gets fake images from the Generator and real images from an image set.
The Discriminator's job is to decide if an image if real or fake.
When the Discriminator is incorrect then the Discriminator will be trained.
When the Discriminator is correct then the Genarator will be trained.
After the Genarator is trained then it will produce better fakes.
After the Discriminator is trained then it will better distinguish fakes from reals.
They are competing with each other and getting better and better at their job.
https://www.youtube.com/watch?v=Sw9r8CL98N0 
