DeepFace

Deepface is one of Facebook developed software for facial recognition , the Artificial Intelligence and Deep Learning used. The goal of DeepFace is to recognize human faces in photos. According to Facebook, the software now achieves an accuracy of 97.35% (deviations of ± 0.25%). This means that DeepFace is almost as good as people who have an accuracy of 97.53%. In addition, DeepFace is a lot better than the FBI software , which only achieves an accuracy of 85%. One of the reasons for this is that Facebook has a lot more pictures to train than the FBI. The software has an artificial neural network with nine layers and 120 million weighted connections. DeepFace was trained with four million images from more than 4,000 Facebook users. According to Facebook, the software should also be used to protect the privacy of users. Facebook wants to check every uploaded image to see whether other Facebook users can be recognized there and then give users the opportunity to make themselves unrecognizable. According to Yaniv Taigman (Artificial Intelligence Scientist at Facebook), the error rate was reduced by over 25% compared to previous software that did the same job. Another special feature of DeepFace is that a 3D model is created so that the person can virtually look into the camera .

history

Facebook introduced facial recognition in Germany in June 2011 . However, since there was a lot of criticism from privacy advocates, Facebook deleted all data the software had collected and deactivated DeepFace across Europe in the fall . In mid-2012, Facebook bought the Israeli company Face.com, which was founded in 2007, for 100 million US dollars because Facebook wanted to expand it. Facebook first presented this project at IEEE 2014.

Web links

DeepFace on Github

Individual evidence

^ Russell Brandom: Why Facebook is beating the FBI at facial recognition. July 7, 2014, accessed May 23, 2020 .
↑ DeepFace: Closing the Gap to Human-Level Performance in Face Verification. Retrieved May 23, 2020 .
↑ ^a ^b Facebook uses the new Deepface technology for its face recognition. Retrieved May 23, 2020 .
↑ Amit Chowdhry: Facebook's DeepFace Software Can Match Faces With 97.25% Accuracy. Retrieved May 23, 2020 (English).
↑ New face recognition system on Facebook - Law, Tax, Economy - CHBECK publishing house. Retrieved May 23, 2020 .
Jump up ↑ Yaniv Taigman, Ming Yang, Marc'Aurelio Ranzato, Lior Wolf: DeepFace: Closing the Gap to Human-Level Performance in Face Verification . In: 2014 IEEE Conference on Computer Vision and Pattern Recognition . June 2014, p. 1701–1708 , doi : 10.1109 / CVPR.2014.220 ( ieee.org [accessed May 23, 2020]).

[1] Russell Brandom: Why Facebook is beating the FBI at facial recognition. July 7, 2014, accessed May 23, 2020 .

[2] DeepFace: Closing the Gap to Human-Level Performance in Face Verification. Retrieved May 23, 2020 .

[:0-3] Facebook uses the new Deepface technology for its face recognition. Retrieved May 23, 2020 .

[4] Amit Chowdhry: Facebook's DeepFace Software Can Match Faces With 97.25% Accuracy. Retrieved May 23, 2020 (English).

[5] New face recognition system on Facebook - Law, Tax, Economy - CHBECK publishing house. Retrieved May 23, 2020 .

[6] Jump up ↑ Yaniv Taigman, Ming Yang, Marc'Aurelio Ranzato, Lior Wolf: DeepFace: Closing the Gap to Human-Level Performance in Face Verification . In: 2014 IEEE Conference on Computer Vision and Pattern Recognition . June 2014, p. 1701–1708 , doi : 10.1109 / CVPR.2014.220 ( ieee.org [accessed May 23, 2020]).