Google has unveiled a new facial recognition system that it claims is the most advanced ever made, according to a Fortune article by Derrick Harris. Called FaceNet, the system was revealed in a new paper published by Google’s three-person research team.
According to the Google researchers, when tested on Labeled Faces in the Wild, a dataset widely used by facial recognition developers, FaceNet achieved 100 percent accuracy. That’s a marked improvement over Facebook’s DeepFace system, which comes close with a 97.25 percent accuracy rate. While DeepFace is already starting to pop up for some Facebook users around the world, helping to tag individuals in users’ photos and so on, the apparently perfect accuracy of Google’s new system could prove valuable in a broader range of applications, from social dating apps to law enforcement investigations.
Facebook’s DeepFace is a product of deep learning, a method of improving artificial intelligence that a number of major companies are starting to explore. Deep learning essentially involves training AI systems to recognize patterns and to ‘learn’ over time, and was recently used by Stanford University researchers to hone their own Deep Dense Face Detector, which they believe is uniquely skilled at detecting faces from a wide range of angles. Harris points out that a number of major organizations are now exploring deep learning systems for their own purposes, so we could soon see rapid advancement in a range of fields, from driverless car systems to consumer robots.
March 20, 2015 – by Alex Perala