There are some limitations: for example, if you remove the eye of a person, Photoshop will not know to replace it with another – but the artificial intelligence of Nvidia knows it. Yes, recently, Nvidia’s researchers have created a technology called “image inpainting” that can reconstruct images with holes. Even if you remove half of a photo, the AI will be able to fill in the rest.
How does it work? An algorithm was trained with 55,116 combinations of original photos and with “mask”, that is, with random holes covering part of the image. In this way, this AI (Artificial Intelligence) learned how to reconstruct the missing pixels. To measure the accuracy of the algorithm, the researchers tested 25,000 masks but did not show the original image. Yes, it means this AI had to rebuild, on its own, the missing parts of the image. You can check the algorithm in action in the video below, in which you can see that it could do better in some situations – the eye it puts on the faces is not convincing – but overall, the result is quite realistic:-
“Our model can handle holes of any shape, location or distance from the edges of the image. Previous approaches focus on rectangular regions located near the center of the image, and often rely on post-processing that requires a lot of resources,” said the researchers at Nvidia in the study. The algorithm ran on Tesla V100 graphics chips and, according to Nvidia, “this AI could be implemented in image editing software to remove unwanted content.” Will does adobe show its interest towards this new AI for its most used Photoshop? So, what do you think about this? Simply share all your views and thoughts in the comment section below.
