Nvidia has unveiled a new state-of-the-art deep learning method that fix or reconstruct corrupted images on the fly, with totally realistic results. In the demonstration, entire chunks of photos can be deleted and Nvidia’s AI will automatically fill in the gaps or missing pixels. It’s handy for corrupted images, but it’s also useful for the reverse effect - deleting unwanted parts from an image and then reconstructing whatever is behind it.
This is wildly tangential to gaming but I thought this was a really neat thing to share anyway as the results are hugely impressive. Nvidia has dubbed it “image inpainting”, clearly with the aim of making this available in photo editing software.
“Our model can robustly handle holes of any shape, size location, or distance from the image borders. Previous deep learning approaches have focused on rectangular regions located around the center of the image, and often rely on expensive post-processing,” the NVIDIA researchers stated in their research paper. “Further, our model gracefully handles holes of increasing size.”
Have a look at for yourself below, it’s quite a party trick.
Nvidia created this tool by training its AI on over 80,000 images with random streaks and holes, grading the results in six categories in order to improve the accuracy of the reconstruction. The results of that work are shown above and, bar a few inescapable oddities, it’s damn near perfect.
“To the best of our knowledge, we are the first to demonstrate the efficacy of deep learning image inpainting models on irregularly shaped holes,” said the NVIDIA researchers.
Now I just need this technology to make its way into Photoshop as there’s evidently lots of fun to be had. It would also pretty much spell the end of watermarks, which is good or bad thing depending on your point of view.
What do you make of Nvidia’s latest tech? Let us know below!