A few days ago we talked about an android app that promises to fix blurry photos, and its performance is reasonably good, although it is not perfect, it depends a lot on the initial image.
Today we present something much more sophisticated, a tool that uses Artificial Intelligence to obtain information that does not exist in the original image.
The tool has been created by researchers at Duke University, and what it does is search through AI-generated images of HD faces, until it finds those that look like the input image. Remember that it is very easy to create fake faces using Artificial Intelligence, we already discussed it in the article. The website makes available 100 thousand images of faces to be downloaded for free.
Photos are created using Generative Adverse Networks (GAN), which produce new images by pitting two neural networks against each other. One of them generates new images and the other checks if they look real. The first network keeps getting better with experience until the second one thinks the images could be real.
The equipment is capable of converting an extremely pixelated face into realistic faces with up to 64 times the resolution of the input. It is clear that the result does not have to be the same as the person who was offered as a model in the first case, but it will be quite similar.
They claim that you can even imagine new features that were unnoticeable in the original, like wrinkles, eyelashes, and so on.
What they do with faces they may do in the future with landscapes, photos taken from satellite, for example, or to see details when zooming in from google maps, for example.
You can see the details at pulse.cs.duke.edu.