top of page

This gif shows how the network gradually learned the features of human faces and generates more realistic ones.

A typical GAN consists of two networks: a generator and a discriminator.

 

In this case, the generator(G) turns random input vector into a image using transpose convolution. The discriminator(D) is a normal CNN that learns to classify fake images against real ones.

 

During training, generator learns to make data that is indistinguishable from real data to the discriminator. After training, we use the generator network alone to synthesize artificial faces.

bottom of page