It's the most instantly irritating but all-too-common social media blunder: your friend uploads a picture where they look amazing but everyone else looks disinterested, unprepared, or has their eyes shut. Now, Facebook may have created a solution that allows your friends to keep up their pictures online without forcing you to subtly untag yourself.
Two Facebook engineers published a paper this week writing that the website may want to use machine learning to fix closed eyes in pictures. An Artificial Intelligence program called a Generative Adversarial Network (GAN) is being developed. It uses images of people with their eyes open to learn what that looks like and then uses "in-painting" to correct images where their eyes are closed taking into account their eye color and shape. The technology is reminiscent of tools that reduce red eye, add makeup, and smooth out skin.
The results are very convincing: Motherboardreports that in one study, people were asked to guess which images contained "in-painted" eyes and which were authentic, and they selected the ones created by Artificial Intelligence 54 percent of the time.
However, there are still a few issues with the technology. It doesn't always register faces at extreme angles, those that are wearing glasses, or those that have hair obstructing the person's eyes. Machine learning gets more and more accurate the larger the size of the data it is given, so researchers feel these problems could be ameliorated if GAN is given more images of eyes to work with.