Deep Learning for Computer Vision

Image translation

An image can be translated to another image, as we have already learned in the application's section. The input images are given to the discriminator, whereas the target images are given to the generator while creating the GAN model as shown here:

gan = tf.contrib.gan.gan_model(    get_generator,    get_discriminator,    real_images,    input_images)

The least square loss is used for training, in addition to the pixel-wise loss to train the model. It can be calculated using the following code:

loss = tf.contrib.gan.gan_loss(    gan,    tf.contrib.gan.losses.least_squares_generator_loss,    tf.contrib.gan.losses.least_squares_discriminator_loss)l1_loss = tf.norm(gan.real_data - gan.generated_data, ord=1)gan_loss = tf.contrib.gan.losses.combine_adversarial_loss( ...