Facebook learned to recognize faces in a crowd with an accuracy of 97.25%

Facebook developed an algorithm called DeepFace, which allows you to identify a person in a crowd with an accuracy of 97.25%, which almost corresponds to the abilities of the average person (97.53%), writes TechCrunch.

To overcome the limitations of conventional face recognition software, Facebook developers have found a way to build 3D models of a face from a photograph. These models can then be rotated to recognize one face captured at different angles. In the past, an attempt to recognize a face could easily become unsuccessful if a person simply tilted his head a little to the other side.

For the full work of the DeepFace algorithm, it must be trained on a large number of individuals. In the current version, it can identify up to 4 thousand people based on a database of over 4 million individual images. Theoretically, this database can be expanded in order to apply to the social network Facebook itself, which would be useful if Facebook wants to automate the process of identifying all your contacts and recognize faces on photos without the need for manual marking by users.
')
So far, this project is only in the form of a research paper published last week, and a group of authors is going to present the results of their work at the Computer Vision and Pattern Recognition conference in Columbus, Ohio, in June. Nevertheless, the project has great potential for future application, both in Facebook itself and in the field of artificial intelligence.