Abstract
Even though SR-GAN was able to make a huge improvement, there was still a gap between the generated image and the ground truth image. The proposed ESR-GAN further enhances the performance.
Three Key Modification Components
- Network
Remove all batch Layer + Replace previous basic building blocks with the proposed Residual in Residual Dense Block. - Change standard discriminator to a Relativistic Discriminator
- Perceptual Loss
Instead of using the feature maps after the activation(which can over-smooth the information) to compare and calculate the loss, ESR-GAN compares the feature maps before the activation.
