Check out this interesting neural network that can restore facial details in a single forward pass.
If you wanted to restore an old, noisy, low-quality image, here's a new neural network that can probably help you to accomplish this task. Meet GFPGAN, a tool for the restoration of facial features created by Xintao Wang, Yu Li, Honglun Zhang, and Ying Shan. According to the team, GFPGAN leverages a rich and diverse library of priors encapsulated in a pre-trained face GAN for blind face restoration. The Generative Facial Prior (GFP) is incorporated into the face restoration process via novel channel-split spatial feature transform layers, which allow their method to achieve a good balance of realness and fidelity.
"Thanks to the powerful generative facial prior and delicate designs, our GFP-GAN could jointly restore facial details and enhance colors with just a single forward pass, while GAN inversion methods require expensive image-specific optimization at inference," says the team. "Extensive experiments show that our method achieves superior performance to the prior art on both synthetic and real-world datasets."