Hands-On Generative Adversarial Networks with PyTorch 1.x

Defining training loss

The loss of SRGAN consists of 4 parts. Here, we let denote the low-resolution (LR) image, denote the super-resolution (SR) image given by the generator, and denote the real high-resolution (HR) image:

• Adversarial loss , as similar to previous GAN models
• Pixel-wise content loss , which is the MSE loss between the SR and ...