Identifying faces is a relatively simple task if you're a human, but it's been a long road for computers to do the same thing. Now Facebook says it's developed a technology for verifying whether two people in side-by-side photos are the same that comes pretty close to replicating human abilities. That project is called DeepFace, and according to Facebook it's 97.25 percent accurate, which is just shy the 97.5 percent humans have scored in the same standardized test. In order to pull off that feat, the technology maps out 3D facial features, then makes a flat model that's filtered by color to characterize specific facial elements. Facebook also says it's tapped into a pool of 4.4 million labeled faces from 4,030 different people on its network in order to help the system learn.
The research project isn't immediately ending up on Facebook. Instead, the MIT Technology Review reports that Facebook's released it ahead of presenting it at the IEEE Conference on Computer Vision and Pattern Recognition this June, all in order to get feedback from the research community.
Facebook introduced facial recognition — that is, the actual capability to figure out who a person is in a photo — in late-2010. The feature was initially available only to US users before the company made it worldwide in 2011, drawing scrutiny in Germany and Ireland where privacy authorities claimed Facebook hadn't given users warning or required consent. The feature also got its fair share of scrutiny from the US, including the ire of Senator Al Franken, who in 2012 grilled the company for not clearly warning users about it. Facebook, which gets more than 350 million photos uploaded by users each day, has pointed to the tool as a way to speed up tagging.