Facebook’s Automatic Tagging Software Has Near-Human Facial Recognition
This is why we need a dislike button.
Bad news for those of us who strategically choose which images of ourselves we label on Facebook. Actually, bad news for anyone who has a face. The social network’s horrifyingly accurate facial recognition software, DeepFace, is approaching human-level accuracy and definitely knows what you look like. NSA, don’t you even think about it.
Of all the social networking sites the name DeepFace could be affiliated with (specifically, I’m thinking DeepFace Cupid and DeepFace Tinder) Facebook actually sounds the least invasive. However, considering the recent revelations about the relationship between Facebook and the NSA (knowing what I wore to my junior prom is apparently vital to the war on terror), forgive me some misgivings over the intimidating new artificial intelligence.
The software uses 3D facial recognition to make identifications from a database of over 4000 individuals, detecting faces with a 97.25 percent accuracy rate. Just for context, our puny human brains only have facial recognition of up to 97.5 percent, so DeepFace could soon surpass our own ability to tag friends and recognize others.
In modern face recognition, the conventional pipeline consists of four stages: detect => align => represent => classify. We revisit both the alignment step and the representation step by employing explicit 3D face modeling in order to apply a piecewise affine transformation, and derive a face representation from a nine-layer deep neural network. This deep network involves more than 120 million parameters using several locally connected layers without weight sharing, rather than the standard convolutional layers.