- This AI turns blurry pixelated photos into hyper-realistic portraits that look like real people. The system automatically increases any image’s resolution up to 64x, imagining features such as pores and eyelashes that weren’t there in the first place.
- The researchers will present their method, called PULSE at the 2020 Conference on Computer Vision and Pattern Recognition (CVPR), held virtually from June 14 to June 19.
The researchers of Duke University have developed a tool based on Artificial Intelligence that can turn blurry, unrecognizable pictures of people into clearly convincing computer-generated portraits. Its features and details are surprisingly finer than ever before. They have come up with a way to take a handful of pixels and generate realistic portrait photos.
Duke computer scientist and the team leader, Cynthia Rudin exclaims that they have never created such super-resolution images at this resolution before this much detail. The system cannot be used to identify people, but rather create new faces that don’t exist but look real.
The Duke team came up with a different approach unlike traditional approaches of taking a lower resolution image and adding new detail, to generate high-resolution faces, with the same size. The tool they have used is ‘generative adversarial network’ (GAN) which has two neural networks trained on the same data sets of photos. Two holds different functions. One comes up with AI-created human faces that mimics the one it was trained on, whereas the other takes the output and make decisions. The first works better and better until the second can’t tell the difference.
From a single blurred image, it can raise a number of uncannily lifelike possibilities, each of which looks like a different person. Even given the pixelated photos, they are able to do something from it, rather than the traditional approaches. The system can convert a picture into higher pixels and quality within seconds. Details such as pores, wrinkles, etc. become crisp and clear in these computer-generated versions.
The customer ratings and scaling generated did the best scoring, almost as high as high-quality photos of actual people.