Monday, November 27, 2017

Face Recognition with 97% Accuracy at 10% of Apple Face ID Power

ESMChina presents Ginwatec Deep Neural Processor that implements face recognition with 97% accuracy at about 10-80x lower power than Apple Face ID. I wonder whether they find a good application for that:


Ark-Invest publishes a grossly incomplete list of Neural Processing chip companies:


Meanwhile, Alipay introduced face recognition-based "Smile to Pay" service in China:

Posted by Vladimir Koifman at 18:24

10 comments:

  1. AnonymousNovember 27, 2017 at 8:09 PM

    Impressive!

    ReplyDelete
  2. bostoncNovember 27, 2017 at 8:27 PM

    Don't understand why China wants to automate away the service jobs. It has a double digit unemployment rate, not to counting many underemployed. The pursuit of cutting edge technologies for the bragging rights doesn't serve the needs.

    ReplyDelete
    Replies
    1. AnonymousNovember 27, 2017 at 11:24 PM

      I think you may forget there is a dot between the double digits.
      https://ig.ft.com/sites/numbers/economies/china

      Delete
  3. AnonymousNovember 28, 2017 at 9:27 AM

    Population shrink due to one-child policy

    ReplyDelete
  4. AnonymousNovember 28, 2017 at 1:20 PM

    So she paid by entering some numbers after smiling, WOW, so NOT impressed, would have been faster to swipe an Apple Watch. :D 2D face authentication is not secure. Those presentations can play with numbers all they want, not even clear if they're referring to 2D+3D face authentication, without a working product... it's fluff.

    ReplyDelete
  5. AnonymousNovember 28, 2017 at 3:26 PM

    From 97% to 99.9999% claimed by Apple FaceID, there is a huge distance !!

    ReplyDelete
  6. AnonymousNovember 28, 2017 at 6:37 PM

    97% success rate means 1 in 33 failure rate. I.e. of 100,000 people, 3000 can pass as you and hack your account. That is the exact opposite of "secure".

    ReplyDelete
  7. AnonymousNovember 28, 2017 at 8:35 PM

    The false positive so far is with identification, ie one to one verification. The rate will go even higher once the recognition mode is enabled.

    ReplyDelete
  8. RobNovember 28, 2017 at 11:54 PM

    Instead of "Smile to pay" I want "Frown to earn".

    ReplyDelete
  9. AnonymousNovember 29, 2017 at 9:03 PM

    Interesting, getting really close to minority report people tracking

    ReplyDelete

All comments are moderated to avoid spam and personal attacks.

Subscribe to: Post Comments (Atom)