Most people touch up their selfies to remove pimples or ruddy patches without a second thought. Facebook wants you to have the same attitude about removing pesky eyelids from photos, too.
Using a machine learning technique called 'generative adversarial network,' or GAN, Facebook researchers taught an AI to observe a picture in which you blinked, compare it to an unblinking photo of you, and then use “in-painting” to substitute your closed eyelids for open eyes.
Facebook researchers Brian Dolhansky and Cristian Canton Ferrer released a paper Monday that detailed the process. Eventually, Facebook could add GAN as a feature to its Facebook or Instagram apps, though the paper does not suggest this is an option.
Dolhansky and Ferrer taught their AI to match features like iris color, skin tone, eye contours and lighting. They then gave photos to participants in their study and asked them to pick which eyes were the originals. 54% of the time, they either picked the in-painted eyes or couldn’t tell the difference.
Dolhansky and Ferrer did admit that they had trouble generating eyes from some original photos, especially those whose subjects wore glasses or whose hair covered or overshadowed their eyes. They’re confident they can solve the issue with further tests.
The study doesn’t make clear whether the “participants” were looking at their own photos or at strangers. People looking at their own selfies or at family snapshots may be more likely to spot an “uncanny valley” effect, as the researchers put it.
Still, with further tweaking, this AI could make it possible to salvage those group photos where one person blinked or everyone squinted in the bright sun.
Facebook isn’t the only tech company interested in using AI to significantly edit people’s features in photos.
Beyond this photo-based plastic surgery, Adobe also uses AI to intelligently select and crop out objects in a photo, or to enhance low-res images by using AI to estimate what pixels are missing and then insert them automatically.
Eventually, AI could be used to selectively pull features from your most flattering photos into some kind of realistic-looking amalgamate selfie, or to pull features from other people’s photos into your own.
Combined with AI tech to simply edit your facial template to look more handsome or beautiful, and we’re about to get an influx of some very flattering, very unrealistic dating profile pics.
Get daily insight, inspiration and deals in your inbox
Get the hottest deals available in your inbox plus news, reviews, opinion, analysis and more from the TechRadar team.
Michael Hicks began his freelance writing career with TechRadar in 2016, covering emerging tech like VR and self-driving cars. Nowadays, he works as a staff editor for Android Central, but still writes occasional TR reviews, how-tos and explainers on phones, tablets, smart home devices, and other tech.