It is common for old photos to be damaged. An open-source artificial intelligence algorithm developed by researchers at Tencent in China promises to restore them in just a few seconds.
The creators report that the system corrects small defects without interfering with the originality of the photo. The resource has a rather technical name: Generative Facial Prior-Generative Adversarial Network (GFP-GAN).
The solution uses a version of Nvidia’s StyleGAN-2 artificial intelligence model. Although the promise is to eliminate noise, correct creases caused by folds, highlight obscure details and improve colors, in tests carried out by the Itatiaia report, such complete results were not obtained. Look!
In the first example, the tool improved the images of the subjects well, but ended up improving the wear and tear on the image, making them clearer. The second photo, with little wear, gained more clarity. These are good results, but you need to know that image wear will not be eliminated.
No addition of elements
Unlike options that enhance photographs, GFP-GAN does not add new elements to the image. It just clears the scene and, when necessary, fills in gaps with information obtained from a database. In addition, artificial intelligence enhances facial details such as eyes, mouth and nose.
The developers inform that there are still limitations and that the model still needs to be improved to avoid distortions that could cause a slight change of identity. The tool also works for low resolution photos and digital images that are corrupted or damaged during capture.
GFP-GAN can be tested for free on the dedicated website or downloaded from GitHub. As it is open source, it can be used in custom tasks integrated with new software. An improved version should be released soon: the expectation is that it will include improved facial recognition, detection of elements with very low resolution and the ability to fill in gaps with a higher hit rate.