Hands-On Generative Adversarial Networks with PyTorch 1.x

Cycle consistency-based model design

Two pairs of generator and discriminator networks are used, with each being responsible for a translation direction. In order to understand why CycleGAN is designed as such, we need to understand how the cycle consistency is constructed.

In the following diagram, the generator, maps sample A to sample B and its performance is measured by the discriminator, . At the same time, another generator, , is trained ...