Fig. 12
From: Probabilistic brain MR image transformation using generative models
The conditional Wasserstein generative adversarial network (cGAN) architecture used in this work. The generator \(\varvec{g}\) receives the input image \(\varvec{x}\), and generates a synthesized image \(\varvec{y}^{\varvec{g}}\) for any random latent vector \(\varvec{z}\). The critic d distinguishes between the generated synthesized image \(\varvec{y}^{\varvec{g}}\) and target image \(\varvec{y}\) paired with \(\varvec{x}\). The training process involves updating \(\varvec{g}\) such that its outputs are indistinguishable from real images by d and updating d to become better at distinguishing the generated images from real ones.
