I’m fan of Facebook’s facial recognition technology and how efficiently is tags you automatically in pictures. Facebook is constantly improving its technology to recognize you and now it doesn’t need to see your face anymore.
acebook’s artificial intelligence team has been working these days on a new experimental facial recognition algorithm that recognizes the people even when their faces aren’t clear (most part covered, looking away etc.). Instead, it looks for other characteristics to recognize the person. For example– it could now recognize you using your hairstyle, clothes, posture, body etc. Awesome but creepy- right?
Yann LeCun, head of artificial intelligence at Facebook challenged himself and his teammates to create a new algorithm that could adapt itself according to the situations when Facebook is unable to see someone’s face clearly. So, this artificial intelligence works just like humans as we can recognize our friends even when their face is covered.
The Facebook AI team took about 40,000 pictures from Flickr with some pictures in which people’s face weren’t fully visible and processed them with the help of a sophisticated neural network. The final algorithm designed was able to recognize the people with covered faces with an excellent 83% accuracy which is better than other facial recognition technologies that make use of clear and uncovered faces.
Facebook has recently launched Moments, a new photo sharing app for mobile and this could be effectively used to power it.
“There are a lot of cues we use. People have characteristic aspects, even if you look at them from the back,” says LeCun. “For example, you can recognise Mark Zuckerberg very easily, because he always wears a gray T-shirt.”
But the privacy implications too won’t be far once this sci-fi tech is launched as the people who don’t want to look at camera would surely raise questions regarding privacy concerns.
So good luck – now you can’t hide from the human-like perfection of Facebook’s facial recognition.
Source: New Scientist
Post a Comment