Facebook DeepFace is 97% Accurate


When you see someone you know, your brain identifies that person’s face in nanoseconds. It is a relatively simple task for us to accomplish.  For a computer, on the other hand, things are not that simple. Cue the Facebook DeepFace project.

The social media firm says it has developed a technology to verify two people in side-by-side photos and see if they are the same person. This technology is so finely tuned that it has been reported to be accurate to 97%.  According to Facebook the project is actually 97.25 percent accurate, which is very close to the 97.5%, which humans have scored, in the same standardized test.

Facebook DeepFace is 97% Accurate - Internet
Facebook DeepFace is 97% Accurate

In order to achieve this type of accuracy, the technology maps out 3D facial features.  Next it makes a flat model that’s filtered by colour to characterize specific elements within the face. Facebook also says the system has access to 4.4 million labelled faces from over 4000 different people on its network to help it learn.

Facebook plans to see what the research community thinks about the technology before it becomes implemented into Facebook. The team at Facebook will be presenting this technology at the IEEE Conference on Computer Vision and Pattern Recognition this June.

Facebook first introduced facial recognition technology in late-2010. The feature was initially accessible only to US users.  The company made it worldwide a year later in 2011.  This drew scrutiny from users in Germany and Ireland where privacy authorities claimed Facebook had not given users warning or asked their prior consent.

Facebook receives uploads of more than 350 million photos each day.  This technological tool will used as a way to speed up tagging.

If you have any sensible comments regarding this story, please leave your comments in the section below.

[Image via f-b]

SOURCE: http://www.theverge.com/2014/3/17/5518808/facebooks-working-on-facial-verification-thats-nearing-human-levels 

Post a Comment