SoylentNews is people
SoylentNews
Arthur T Knackerbracket has found the following story:
Hanwang, the facial-recognition company that has placed 2 million of its cameras at entrance gates across the world, started preparing for the coronavirus in early January.
Huang Lei, the company’s chief technical officer, said that even before the new virus was widely known about, he had begun to get requests from hospitals at the centre of the outbreak in Hubei province to update its software to recognise nurses wearing masks.
[...] If three or five clients ask for the same thing . . . we’ll see that as important,” said Mr Huang, adding that its cameras previously only recognised people in masks half the time, compared with 99.5 percent accuracy for a full face image.
[...] The company now says its masked facial recognition program has reached 95 percent accuracy in lab tests, and even claims that it is more accurate in real life, where its cameras take multiple photos of a person if the first attempt to identify them fails.
“The problem of masked facial recognition is not new, but belongs to the family of facial recognition with occlusion,” Mr Huang said, adding that his company had first encountered similar issues with people with beards in Turkey and Pakistan, as well as with northern Chinese customers wearing winter clothing that covered their ears and face.
Counter-intuitively, training facial recognition algorithms to recognize masked faces involves throwing data away. A team at the University of Bradford published a study last year showing they could train a facial recognition program to accurately recognize half-faces by deleting parts of the photos they used to train the software.
(Score: 2) by ilsa on Sunday March 22 2020, @01:03PM (1 child)
While I have knowledge of the details of this system, that one statement alone makes me call shenanigans. I wouldn't be surprised if the accuracy is a good 25% lower than stated, if not lower still.
There is no way you can implement a system like this without adequate testing*, and you certainly don't just go from single- to multiple-image based recognition willy nilly at the drop of a hat. It's exceedingly unlikely that RL performed better than lab where you can do a much better job of controlling the variables.
He either has no idea how his own system works, or he's lying through his teeth.
*I mean, you can, if integrity and quality arn't a priority.
(Score: 0) by Anonymous Coward on Sunday March 22 2020, @07:00PM
Um. If the lab's train/test has single snapshots in the tests, I can assure you that improved geometries and textures are retrieved from (multiple stills extracted from) video, assuming state of the art. The counterclaim would be much more extraordinary.
Of course phrasing it as "in real life is better than in the lab" is disingenuous; they should be using video not stills in the lab (train and) test sets.