Hands-On Generative Adversarial Networks with Keras

Generator

The generator model consists of a dense layer followed by ResNet blocks and a 2D Convolution layer with the Softmax non-linearity. The Softmax non-linearity normalizes the logits of the last layer to a probability distribution. Remember that, during training, no sampling step is required and the Softmax output is passed directly to the Discriminator:

from keras.layers import Input, Activation, Softmaxfrom keras.layers import Dense, Conv2Dfrom keras.layers.core import Reshapefrom keras.models import Modeldef build_resnet_generator(input_shape, n_filters, n_residual_blocks,                           seq_len, vocabulary_size):    inputs = Input(shape=input_shape)        # Dense 1: 1 x seq_len x n_filters    x = Dense(1 * seq_len * n_filters, input_shape=input_shape)(inputs) ...