This AI can restore your damaged and ripped photos with surprising accuracy

Alex Lee April 25, 2018

The ol’ ‘decayed photo’ is often used as a hokey MacGuffin in numerous films but it’s not uncommon for physical photos to become easily damaged, torn or decayed, especially those taken before the advent of digital cameras and cloud storage.

Until now, the only option was to pay for costly photo restoration services.

To solve this problem, Nvidia has created a way to restore decade-old memories that you’ve already lost using the beauty of AI. According to the company, if there are any missing pixels or holes in a photo, the recently-developed AI will be able to fill it in using a process called “image inpainting”.

Nvidia trained the AI by feeding it more than 50,000 photos which had holes placed in different positions. Using this data set, the AI was taught to recognise and reconstruct the missing areas. If someone in a photo has a missing nose or a missing eye, for example, the AI will be able to fill it in with a computer-generated one.

Unlike the content-aware tool available in Photoshop, which fills the selection with a blend of surrounding pixels, the AI analyses the whole photo to determine what the subject should look like.

“Our model can robustly handle holes of any shape, size location or distance from the image borders,” the Nvidia team wrote in the research paper. “Previous deep learning approaches have focused on rectangular regions located around the centre of the image and often rely on expensive post-processing.”

It isn’t perfect, however. While the AI can’t input the exact same nose as the original photo, it will come scarily close. Similarly, if the photo is damaged beyond repair and there’s a giant hole where an entire person should be, the AI is unlikely to be able to do very much.

That said, Nvidia’s solution provides a very simple way of restoring your photos, which wouldn’t take much effort beyond running it through the tool, and gives us a good indicator of how easy photo editing might be in the future.

For now, there’s no word on when, or even if, we will be able to use the AI, but we’ve reached out to Nvidia to see if it has any plans to make the tool more widely available.