Facebook released the DeepFace system in 2015. The project emerged from a group of scientists including Yaniv Taigman and Ming Yang who joined Facebook after the company acquired Face.com in 2012.
DeepFace achieved 97.35% accuracy on the Labeled Faces in the Wild dataset while human beings reached 97.53% on the same test set. This made the system nearly as effective as people but less accurate than Google's later FaceNet which recorded 99.63%.
Illinois passed the Biometric Information Privacy Act requiring written release and notice for collection because users alleged Facebook violated this law by using facial identification without consent. A class action lawsuit led to courts ordering payment in early March 2021 where sixteen hundred thousand residents received at least $345 each.
The system employs a nine-layer neural network containing over 120 million connection weights that processes images through four distinct modules. These modules include 2D alignment detecting six fiducial points, 3D alignment using 67 anchor points, frontalization warping the image forward, and a convolutional layer followed by max pooling and locally connected layers.
Meta deleted face scan data for over one billion users by December 2021 following plans to shut down Facebook's facial recognition system entirely. This action represented one of the largest shifts in facial recognition usage history while the company stated it had not ruled out incorporating similar technology into future products.