The adversarial loss is calculated on the probabilities returned by the discriminator network. In the adversarial model, the discriminator network is fed with generated images, which are generated by the generated network. The adversarial loss can be represented by an equation, as follows:

Here,  is the generated image and  represents the probability that the generated image is a real image.

The perceptual loss function is a weighted sum of the content loss and the adversarial loss, which is represented as the following equation:

Here, the total perceptual loss is represented by .  is the content loss, which can be either a pixel-wise MSE loss or VGG loss.

By minimizing the perceptual loss value, the generator network tries to fool the discriminator. As the value of the perceptual loss decreases, the generator network starts generating more realistic images.

Let's now start working on the project.