You may not find it difficult to identify a face in side-by-side photos but for computers, this has not proven to be a simple task so far. Now Facebook has come up with new software called DeepFace, which can verify whether two unfamiliar photos of faces show the same person. With 97.25 percent accuracy, the software comes pretty close to replicating human abilities.
It is interesting to note here that humans have scored 97.5 percent in the same standardized test.
DeepFace shows a significant advancement over previous face-matching software by mapping out 3D facial features and then making a flat model that’s filtered by color to characterize specific facial elements. Facebook helped the system learn by tapped into a pool of 4.4 million labeled faces belonging to more than 4,000 different people on its network.
“You normally don’t see that sort of improvement,” says Yaniv Taigman, a member of Facebook’s AI team (a research group created in 2013).
“We closely approach human performance,” adds Taigman who developed DeepFace along with Facebook colleagues Ming Yang and Marc’Aurelio Ranzato and Tel Aviv University professor Lior Wolf.
Facebook first introduced the facial recognition feature in late-2010. It was initially available only to US users until its worldwide launch in 2011.
[Source: MIT Technology Review]