Facebook's DeepFace Facial Matching System Is Almost Perfect

It can tell if two faces are the same person 97.25% of the time

facebook deepface

facebook deepface

If you're paging through a big set of photos, you probably don't have much trouble spotting if the same person pops up multiple times. In fact most humans can match two faces 97.55% of the time. Now Facebook's DeepFace technology has almost matched this, correctly recognizing the same features 97.25%—within spitting distance of human performance.

Keep in mind, this technology isn't facial recognition (putting a name to a face), but facial verification (matching two faces together). But given Facebook's unparalleled collection of photos tagged with your identity, moving from one to the other wouldn't be hard.

DeepFace works by looking at a photo of your face, then correcting the angle of it using a 3D model of an "average face". From there it can identify key features and a "description" of the face, which can be compared to other, similarly analyzed images.

For now, DeepFace isn't coming to your Facebook timeline. As the Technology Review reports:

However, DeepFace remains purely a research project for now. Facebook released a research paper on the project last week, and the researchers will present the work at the IEEE Conference on Computer Vision and Pattern Recognition in June. “We are publishing our results to get feedback from the research community,” says [AI team member Yaniv Taigman], who developed DeepFace along with Facebook colleagues Ming Yang and Marc’Aurelio Ranzato and Tel Aviv University professor Lior Wolf.

But sooner or later, you can be sure it'll transition to helping you tag photos—assuming the intense processing power behind it can be tamed to the point where it can be used to crunch the numbers on the obscene volume of images that get uploaded to Facebook daily.

[via Gawker]