Facebook Develops AI Program That Fixes Photos with Blinking Subjects

The use of artificial intelligence extends to fixing photos where subjects, by accident, have their eyes closed, and a Facebook research aims to introduce a program that can produce realistic results.

Facebook engineers Brian Dolhansky and Cristian Canton Ferrer explain in a paper how an AI-powered program can help reconstruct a photo specifically with subjects who accidentally blinked at the same the image was captured.

The base of the study references the use of Generative Adversarial Networks (GANs), which is practically a deep machine learning network that can produce images of various subjects, such as a human face, even from scratch. So the process of in-painting — or filling in lost segments of an image — should be an easier task for this program.

The Facebook engineers’ study is focused on introducing a program variant they call the Exemplar GANs (ExGANs), where the machine uses extra or “exemplar” information that “corresponds directly to some identifying traits of the entity of interest.”

“We propose a general framework for incorporating this extra exemplar information. As a direct application, we show that using guided examples when training GANs to perform eye in-painting produces photo-realistic, identity-preserving results,” the researchers further explain.

Meanwhile, while the research seems promising, this is not the first time that the same concept of fixing blinking selfies has been offered to photo-loving people. Peta Pixel noted that the Adobe Photoshop Elements 2018 introduced a function called “Open Eyes,” which practically copies the subject’s eye from another image where they are not blinking.

However, as the Facebook engineers show in the paper, ExGAN-generated photos are far more accurate and realistic-looking than the results they got from Adobe’s Open Eyes.

This technology, considering its demonstrated accuracy, is promising. However, there are some questions of whether or not Facebook will actually add this as a feature to the social media platform. The Verge pointed out that an AI that deliberately alters a person’s photo might be against Facebook's goals of letting people express their true selves.