Facebook can fix your blinking selfies with an eerie level of accuracy

If you’re one of those that are doomed to perpetually blink in photos, Facebook has a solution. As long as it has a couple of reference photos where your eyes are fully open, Facebook AI researchers have found that they can fix blinking selfies.

Facebook calls the technique, outlined in a research paper published Monday, “eye-in painting.” Its method incorporates a type of machine learning called exemplar generative adversarial networks, which has proven adept at a variety of image generation tasks. This includes developing fake celebrity faces and designing clothing, among other applications. With ExGANs, Facebook can create “photo-realistic personalized in-painting results” in photos where the subject goes from closed to open eye.

Blinking people Facebook

As you can see in the image above, the system needs at least one photo of the subject with eyes open to function properly. However, with that one photo, it can then successfully edit open eyes onto an image where the subject’s eyes are closed. The system isn’t perfect, however. It has issues when an individual’s face is at a sharp angle, if they are wearing glasses, or if they have long hair covering their eyes.

Still, the results of Facebook’s ExGAN-based eye-in painting are quite good. While the results are photo-realistic, the images do seem to have something lacking, however. Perhaps it’s the lack of facial expression from the closed-eye shot, but the eyes in the AI-generated images seem a little dead. They’re missing the sparkle and personality you can see in the open-eye images of these subjects.

Even so, given a set of shots with the same lighting conditions and facial expressions, it’d be interesting to see if humans could tell the difference between the true eyes-open and AI-generated images, particularly when it’s further perfected.

For now, this technique to fix blinking selfies is merely something Facebook researchers have explored and continue to develop. It’s not something you can use on your Facebook or Instagram photos—yet, at least.


H/T The Verge

Christina Bonnington

Christina Bonnington

Christina Bonnington is a tech reporter who specializes in consumer gadgets, apps, and the trends shaping the technology industry. Her work has also appeared in Gizmodo, Wired, Refinery29, Slate, Bicycling, and Outside Magazine. She is based in the San Francisco Bay Area and has a background in electrical engineering.