Wednesday, November 6, 2013

My answer on facial detection sex accuracy is answered

(The usual disclaimer applies. Although I do not work with face detection software, I do work in the biometrics industry.)

In a previous post, I blogged about the Amscreen technology that Tesco is using for several purposes, including determination of the sex of the person who is viewing an onscreen ad. I wondered about the accuracy of the technology.

It turns out that Amscreen's facial detection technology is provided by a company called Quividi, and in a December blog post, Quividi provided an answer to that very question.

A test was performed by an independent research company in the USA to evaluate the accuracy of Quividi’s classification algorithms.

Viewers saw a short video clip and at the end of that clip the system ‘read’ them and showed a male video clip to males and a female video clip to females.

The test was very successful and overall we saw a 93% “instant” accuracy rate (i.e. the ability to tell for each single person his or her precise gender) after scanning 856 people.


Specifics are available here, including this interesting tidbit:

52 of the females were wearing baseball caps and our accuracy of this group was only 42%. If we look at the accuracy of the software without any women wearing hats, the accuracy rate increases to over 96%

So when you removed the baseball caps, the female sex detection software was in a league of their own. But shouldn't it be the other way around?
blog comments powered by Disqus