Hands-On Generative Adversarial Networks with Keras

GANs

We wrap the generator and discriminator methods that are to be used during training in the following function:

def GAN(G, D, z_dim, n_labels, resolution, n_channels):    # create D(G(z)) and set current block    G_train = Sequential([G, D])    G_train.cur_block = G.cur_block    # instantiate symbolic variables for inputs    shape = D.get_input_shape_at(0)[1:]    gen_input, real_input = Input(shape), Input(shape)    interpolation = Input(shape)    # define the wasserstein distance and grad norm    w_distance = WassersteinDistance()([D(gen_input), D(real_input)])    g_norm = GradNorm()([D(interpolation), interpolation])        # use w_distance, g_norm and output of D on reals for computing the loss    D_train = Model([real_input, gen_input, interpolation], [w_distance, g_norm, ...