The purpose of this repo is to understand better the paper and do also some experiments with code.
The paper name is "๐ป๐๐๐๐๐ ๐ ๐น๐๐๐-๐พ๐๐๐๐ ๐ฉ๐๐๐๐ ๐ญ๐๐๐ ๐น๐๐๐๐๐๐๐๐๐๐ ๐๐๐๐ ๐ฎ๐๐๐๐๐๐๐๐๐ ๐ญ๐๐๐๐๐ ๐ท๐๐๐๐". It was published on 11 Jan 2021 by: Xintao Wang, Yu Li, Honglun Zhang and Ying Shan.
This paper aims at recovering high-quality faces from the low-quality counterparts suffering from unknown degradation, such as low-resolution, noise, blur, compression artifacts etc.
GFP-GAN is comprised of a degradation removal module (U-Net) and a pretrained face GAN. To train the model it uses in total three losses: Adversarial Loss, Facial Component Loss and Identity Preserving Loss.