Facebook researchers have been developing a machine learning technology that can add eyes to a selfie in which the subject is blinking. The company detailed the technology on its Research website recently, explaining that in-painting is used to remove the closed eyelids and replace them with what appears to be the subject’s own eyes.
Facebook uses artificial intelligence for a variety of features and purposes, and one could soon be improving images in which your eyes are closed. Yes, if you’re taking a selfie yourself, you’ll likely just retake the image. However, if you’re in a selfie with someone else — or simply got caught off guard by a camera — this tech could make fix the simple mistake.
According to a research paper published by Facebook this month, the system utilizes a generative adversarial network, more commonly called a GAN. With the technology, machine learning can learn the user’s eyes, then use that data to fill in the blanks when the person blinks. Because the system already knows what their eyes look like, the results look true to life.
Assuming the subject’s face isn’t obscured by hair, their eyes aren’t behind glasses, and similar things, the system appears to do an excellent job of imitating the real thing. This is somewhat like an extension of existing photography technology that does things like remove red eyes from images.
As with any research effort, it’s possible that Facebook won’t implement this into an actual, final product. It’s also possible that we could one day see this introduced in a Facebook product, potentially on Instagram where it would arguably be most useful. The full paper can be accessed here.