Hands-On Generative Adversarial Networks with Keras

The generator

The generator that is used in Progressive Growing of GANs consists of stacks of convolutions followed by an upsampling layer. Each convolution has a WeightScalingLayer layer that normalizes the layer outputs with a constant, and a PixelNormLayer layer that normalizes the outputs by their L2 norm, thus ensuring that the outputs vector has unit length.

Specific to the Progressive Growing of GANs methodology, this generator architecture has a BlockSelectionLayer layer, which defines the layer that should be used as the output during training. Remember, in this methodology, we first train an output layer at a low resolution and then train the other layers that output a higher resolution one by one:

def Generator(n_channels=1, resolution=32, ...