Thanks to artificial intelligence, it is increasingly difficult for a poor resolution to ruin a good photo. And it is that technology advances by leaps and bounds in the generation of images. From generating a similar image, to coloring and going through increased resolution. In fact, this is something Google’s AI does superbly well, capable of convert fully pixelated photos to high resolution photos.
To do this, the company’s machine learning model has learned to take a photo with hardly any resolution and scale it to get unique details. The model used by Google in particular is one called diffusion models, which is generative. In other words, the system takes a low-resolution image as input and from there builds a high-resolution image on its own. Saying it a little more clearly: what is missing is invented but with great realism.
Interestingly, Google has explained that first they have trained the AI to lower the resolution of the images and make them extremely pixelated. From there “learn to reverse the process, starting with pure noise.” In this way he manages to improve, above all, the portraits of people. However, with the intervention of a second AI it is able to scale the photographs even more and at 32 x 32 px resolutions it is capable of reaching an impressive 1024 x 1024 px.
The result are Detailed photos with minimal errors that can pass for real without any problem, especially so that you do not know the context in which they have been taken. This is a very interesting solution for a time when our memories have become almost exclusively digitally immortalized. So we hope to be able to have this technology at the touch of a finger on our smartphones very soon.
Photos | Unsplash and Google