Legendary cinematic intelligence tool that allows a blurred image to get a detailed photo, just became so unbelievably close to reality. Researchers from Duke University (USA) have developed an algorithm PULSE, which is of much pixelization pictures creates photo-realistic portrait. Engaged in this special artificial intelligence.
This AI is a kind of generative adversarial networks, trained it is portraits of people. One neural network uses the pixel image as a set of initial conditions and generates pseudopotto, the second evaluates the result and provides edits.
The algorithm impresses with its features – its authors promise a 64-fold increase in image detail. The picture in the resolution of 16×16 pixels, from which only and it is clear that this kind of a person turns into an image with a resolution of 1024 x 1024 pixels. It is easy to confuse with the real, and the network even cares about such trifles as the pattern on the collar, if it is appropriate.
Due to the fact that the program does not recreate the original image, only generates a perfect copy, it will be difficult to apply for serious problems. On the other hand, the algorithm can be optimized to recreate the look of distant exoplanets, not scientific, and artistic purposes. Or for the visualization of the microcosm to show bacteria and viruses in such a way that they become understandable and interesting not only to specialists but also to students.
Source — Duke University