Not all selfies are bound to look right, as you might often notice you have closed your eyes at the wrong moment. However, Facebook found a way to fix this by using, of course, artificial intelligence. By using such an algorithm, the platform might fix blinking selfies by replacing your closed eyes with a pair of open eyes.
Facebook found a clever way to fix blinking selfies
Blinking selfies are a common problem, especially if you’re not ready for the photograph. When you’re not willing to take a new selfie, Facebook comes up with a solution. It has trained an AI to flag blinking selfies and then ‘open’ your eyes. While this sounds great, you might find a few problems with it.
It’s not uncommon for AIs to fix photos. However, when they do that, they usually replace the wrongs in your photo with parts of a different person’s face. Of course, the results don’t look too good, as the new faces look like unnatural combinations. Fortunately, Facebook surpassed these difficulties.
This algorithm uses photos of the same person to fix the blinking eyes
To avoid creating strange hybrids, the Facebook scientists found a more advanced way to train the AI. Instead of using random photos to replace eyes in blinking selfies, the algorithm will use older photos of yourself. This means it is advanced enough to identify the person in the photo and find other pictures of them.
This sounds great if you don’t like taking multiple selfies, but also puts some worrying prospects. Facebook uses two different methods to do the eye replacement. For the first one, it uses its image database to find a picture of yourself and replace the eyes. For the second one, it stores actual data on your eyes, similar to the facial recognition technique.
Facebook’s algorithm is better at fixing blinking selfies because it developed a different database for it. Usually, most AIs use already existent data. However, the social media giant created a special database for it, with at least three selfies of 200,000 people. Scientists also developed a study on the technique, published on the Facebook Research web page.
Image source: Pexels