Slashdot Mirror

← Back to Stories (view on slashdot.org)

Facebook's Face Identification Project Is Accurate 97.25% of the Time

Posted by Soulskill on from the almost-as-good-as-the-NSA's-version dept.

kc123 tips news that 'DeepFace,' the software research project created by Facebook engineers to identify people in pictures, is now accurate 97.25% of the time. In other words, it's almost as good at recognizing faces as humans, who are able to determine whether two photos show the same person 97.53% of the time. The article says DeepFace reaches that level of accuracy "regardless of variations in lighting or whether the person in the picture is directly facing the camera." It continues, "DeepFace processes images of faces in two steps. First it corrects the angle of a face so that the person in the picture faces forward, using a 3-D model of an 'average' forward-looking face. Then the deep learning comes in as a simulated neural network works out a numerical description of the reoriented face. If DeepFace comes up with similar enough descriptions from two different images, it decides they must show the same face. ... The deep-learning part of DeepFace consists of nine layers of simple simulated neurons, with more than 120 million connections between them. To train that network, Facebook’s researchers tapped a tiny slice of data from their company’s hoard of user images—four million photos of faces belonging to almost 4,000 people."

2 of 149 comments (clear)

  1. Re:Say goodbye by StripedCow · · Score: 5, Interesting

    Muslims are right: Burqa's are the way to go...

    --
    If Pandora's box is destined to be opened, *I* want to be the one to open it.
  2. Training the data by MobyDisk · · Score: 5, Interesting

    I logged into Facebook for the first time in about 6 months, and it required me to authenticate myself by answering a series of questions about who was in each picture. It would display 3 pictures, each showing a square around a particular person, and it would ask who the person is. It was multiple choice.

    I wonder if this is how they confirm that the data is correct, to eliminate intentional errors. You can ask a person who doesn't own the picture and didn't tag it to confirm the person in there. By masking it as an authorization request you convince people who otherwise would not be involved in tagging to participate.