16.03.2021 Views

Advanced Deep Learning with Keras

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Chapter 7

Figure 7.1.15: The backward cycle of CycleGAN with PatchGAN on MNIST (source) to SVHN (target).

The reconstructed target is not entirely similar to the original target. Original color photo can be found on

the book GitHub repository, https://github.com/PacktPublishing/Advanced-Deep-Learning-with-Keras/

blob/master/chapter7-cross-domain-gan/README.md.

In Figure 7.1.3, CycleGAN is described to be cycle consistent. In other words, given

source x, CycleGAN reconstructs the source in the forward cycle as x ' . In addition,

given target y, CycleGAN reconstructs the target in the backward cycle as y ' .

[ 233 ]

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!