Google: Our New System For Recognizing Faces Is the Best

schwit1 writes Last week, a trio of Google researchers published a paper on a new artificial intelligence system dubbed FaceNet that it claims represents the most accurate approach yet to recognizing human faces. FaceNet achieved nearly 100-percent accuracy on a popular facial-recognition dataset called Labeled Faces in the Wild, which includes more than 13,000 pictures of faces from across the web. Trained on a massive 260-million-image dataset, FaceNet performed with better than 86 percent accuracy.

The approach Google's researchers took goes beyond simply verifying whether two faces are the same. Its system can also put a name to a face—classic facial recognition—and even present collections of faces that look the most similar or the most distinct. Every advance in facial recognition makes me think of Paul Theroux's dystopian Ozone.

Comments

Google

[Anonymous Coward]

The NSA and CIA must love the direction this company has taken.

Re:Google

[TheGratefulNet]

there was speculation that, early on, the nsa was a key funder of google. I think there was some dns registration stuff that made people do a double-take (long time ago, when google first started).

could google have gotton so far without nsa's help? one wonders. and one will never actually know, either.

Re:Google

[mbkennel]

> could google have gotton so far without nsa's help? one wonders. and one will never actually know, either.

It's the other way around. NSA was interested in techniques and technology from google, especially high-performance large scale data processing. NSA was/is behind, and they knew it, and they knew the best didn't want to work with them any more when they could get a pre-IPO position at Google when Google had stunningly capable & ambitious people (2000-2005) on average.

Re:Confusion

[RingDev]

The part that intrigues me is that they claim to return a name with the face.

This would imply that their facial recognition isn't just a image match, but that it looks at the context of the photos it finds to attempt to identify meta data about the people within it. Assuming that their facial recognition is no better than anyone else's recognition, by adding meta data to the calculation, especially given Google's propensity to collect and search meta data, it would seem likely that they use the meta data to make stronger identifications and find more reference photos of potential matches.

For example, if they do the first facial only search and come up with 10,000 possible matches, then they do meta searches on those 10,000 to find more pictures of them, then those pictures are compared for stronger 'training', you wind up with a much higher level of accuracy.

-Rick