Hands-On Generative Adversarial Networks with Keras

Auxiliary training functions

We need to implement several functions to accommodate the Wasserstein GAN with Gradient Penalty loss. Our implementation is an adaptation of the implementation written by the Keras team; it can be found at https://github.com/keras-team/keras-contrib/blob/master/examples/improved_wgan.py

We start with the necessary imports:

import keras.backend as Kfrom keras.layers.merge import _Merge

We first define the Wasserstein loss, which is just the average of the Discriminator's output multiplied by the respective sign:

def loss_wasserstein(y_true, y_pred):     return K.mean(y_true * y_pred)

Next, we define the Gradient Penalty loss:

 def loss_gradient_penalty(y_true, y_pred, averaged_samples,                           gradient_penalty_weight): # get ...