Facebook learned to recognize people without a face

Facebook face recognition technologies have reached such a level that they are able to identify a person even if the face is not visible at all (for example, from the back). This was made possible thanks to the latest advances in computer vision systems, writes New Scientist.

The experimental neural network, developed in the AI ​​laboratory, is trained on additional features, such as a hairstyle, hair color, clothes, figure and pose.

Conventional face recognition systems are already working quite well, they are on par with a person, and some even outperformed a person. Facebook decided to go further and try to teach the program to recognize people, even if the face is poorly visible or not visible at all, because people do a good job of this task, explains Yann LeCun, the head of Artificial Intelligence on Facebook.

In fact, even looking from the back there are quite a few signs by which we recognize a familiar person. Say Mark Zuckerberg is easy to recognize because he always wears the same t-shirts.

The neural network was trained in 40,000 photos from Flickr, in some of which the face was clearly visible, and in others - poorly. After training, the system showed a recognition accuracy of 83%.

The results of the study were presented earlier this month at a conference on computer vision and pattern recognition in Boston.

In the future, the algorithm may be implemented in commercial Facebook products. For example, in the Moments program , which automatically sortes the archive of photos by individual events, with recognition of the names of all friends in the photos.

So sunglasses and hoods will no longer hide you from Big Brother,% username%.