Tencent developers have introduced the GFP-GAN artificial intelligence (AI) system, which has the ability to restore old damaged photos, as well as increase their resolution, Engadget reports.
It is known that AI “completes” the missing details in a few seconds, while at the same time maintaining the image quality. The developer pointed out that usually old photos are restored using AI models that look for differences between artificially generated and real images. This often leads to a poor-quality result.
The new approach is to use the existing pre-trained StyleGAN-2 model from Nvidia, which “guided” the developers’ model through several stages of the image creation process. This technique allows you to preserve the “personality” of the person in the photo, as it focuses on such facial objects as eyes and mouth. Tencent noted that while the AI is not working perfectly, and there may still be inaccuracies in the restored image, however, the work done already looks promising.
