It’s not exactly surprising to anyone, but Facebook has been working on facial recognition technologies for years, trying to figure out how to automatically tag users in photographs. Now, however, it looks like the company’s research has reached a new high-point where the technology is approaching “human-level performance.”
What does this mean? Well, it means that the company’s algorithm is so good that it could identify someone better than you can.
Dubbed DeepFace, the facial recognition software maps 3D facial features and turns them into a flat model. It then starts filtering the results by color to identify various facial elements that make people unique.
According to a file released by Facebook, in a test, DeepFace managed to give the correct answer in 97.25 percent of cases when asked to figure out if two people in side-by-side images are the same individuals or not.
Taking the same test, humans got a 97.5 percent score, which backs up Facebook’s statement that DeepFace was approaching “human-level performance.”
The software is still some months away from getting integrated into the social network and it will certainly raise a few eyebrows from privacy advocates, as all facial-recognition tools do.
Hopefully, Facebook will find a way to implement DeepFace without stepping on too many toes.