The facial recognition system developed by Facebook was called DeepFace.
DeepFace was created by researchers at Facebook’s artificial-intelligence team and described in a 2014 research paper. Its approach used 3D face alignment followed by a deep neural network, helping the system compare faces even when photos were taken from different angles or under different lighting conditions.
The reported results were notable for the time: DeepFace reached about 97.35% accuracy on the Labeled Faces in the Wild benchmark, close to the reported human performance of 97.53%. The system was trained using roughly four million images uploaded by Facebook users and used a nine-layer neural network with more than 120 million connection weights.
DeepFace is not the same as FaceNet, a Google-developed system, or DeepFaceLab, an unrelated open-source face-swapping project. Facebook later ended its facial-recognition feature for users, but DeepFace remains the name of the underlying research system.