Hands-On Convolutional Neural Networks with TensorFlow

Putting the losses together

In TensorFlow, we can implement the whole GAN loss, as shown in the following code. As input, we take the output of the discriminator for a batch of fake images from the generator and a batch of real images from our dataset:

def gan_loss(logits_real, logits_fake):         # Target label vectors for generator and discriminator losses.     true_labels = tf.ones_like(logits_real)     fake_labels = tf.zeros_like(logits_fake)     # DISCRIMINATOR loss has 2 parts: how well it classifies real images and how well it     # classifies fake images.     real_image_loss = tf.nn.sigmoid_cross_entropy_with_logits(logits=logits_real, labels=true_labels)     fake_image_loss = tf.nn.sigmoid_cross_entropy_with_logits(logits=logits_fake, labels=fake_labels) # Combine ...