Deep Learning for Computer Vision

Vanilla GAN

The original GAN is called a vanilla GAN. Before we construct the model, let's define a few layers that will be useful for the rest of the chapter. The following is the convolutional_layers with leaky activations and regularization added:

def convolution_layer(input_layer,                      filters,                      kernel_size=[4, 4],                      activation=tf.nn.leaky_relu):    layer = tf.layers.conv2d(        inputs=input_layer,        filters=filters,        kernel_size=kernel_size,        activation=activation,        kernel_regularizer=tf.nn.l2_loss,        bias_regularizer=tf.nn.l2_loss,    )    add_variable_summary(layer, 'convolution')    return layer

Next, we will define a transpose_convolution_layer that is the opposite of a convolution_layer with regularization, using the following code:

def transpose_convolution_layer ...