Hands-On Generative Adversarial Networks with Keras

Discriminator

The Discriminator model consists of a Convolution followed by ResNet blocks and a dense layer that projects the learned features to a single number, as given in the following code block:

from keras.layers import Input, Flattenfrom keras.layers import Dense, Conv2Dfrom keras.models import Modeldef build_resnet_discriminator(input_shape, n_filters, n_residual_blocks,                               kernel_size=(1, 1), stride=1):    input = Input(shape=input_shape)    x = Conv2D(filters=n_filters, kernel_size=kernel_size, strides=stride,               padding='same')(input)    # resnet blocks    x = resnet_block(x, n_residual_blocks, n_filters)    # Flatten and project to a single dimension    x = Flatten()(x)    x = Dense(1)(x)        # create model graph    model = Model(inputs=input, outputs=x, ...