Facebook’s 2014 deep-learning facial-recognition system was called DeepFace. The system used a deep neural network to produce face representations and compare them for identity recognition.
The DeepFace research team trained the model on a very large collection of face images and used a three-dimensional alignment step to reduce differences caused by pose. The model was reported to achieve near-human performance on the Labeled Faces in the Wild benchmark used in the study.
DeepFace is often confused with FaceNet, a later system developed at Google that also learned compact face embeddings. It is also distinct from DeepMind, the British artificial-intelligence company acquired by Google. Facebook later rebranded as Meta, while the DeepFace name remains associated with the 2014 research paper.