What’s funny is this was all predictable.
We had the same problem when we started with photography in general. It worked well on white people but made every black person a featureless black void.
Turns out we test these things mainly on white men then call it a success when it works exclusively for us white men.
Uh? There has never been a time when it was not practical to photograph black people, since photography first became practical. Some autoexposure algorithms are shit at black people (for the reason you give), but that came long after its inception and doesnt actually prevent taking a good picture.
Over a decade ago I was hired as a third party to a third party to a company that was hired by some confidential customer we would never get the identity of. There was some biometric stuff involved on the project, to which we had to import some devices - but they got stuck on customs for a while simply because it had bio in the name. We tried to make do with what we could and were pretty confident when we tested with the real device on the last day before a supposed presentation we also knew nothing about.
Next day we get the news: it went terrible. And we got a feedback: “it failed to identify most kids on the school”. That’s fair, we never even considered testing it with kids.
What’s funny is this was all predictable. We had the same problem when we started with photography in general. It worked well on white people but made every black person a featureless black void.
Turns out we test these things mainly on white men then call it a success when it works exclusively for us white men.
Uh? There has never been a time when it was not practical to photograph black people, since photography first became practical. Some autoexposure algorithms are shit at black people (for the reason you give), but that came long after its inception and doesnt actually prevent taking a good picture.
Over a decade ago I was hired as a third party to a third party to a company that was hired by some confidential customer we would never get the identity of. There was some biometric stuff involved on the project, to which we had to import some devices - but they got stuck on customs for a while simply because it had bio in the name. We tried to make do with what we could and were pretty confident when we tested with the real device on the last day before a supposed presentation we also knew nothing about.
Next day we get the news: it went terrible. And we got a feedback: “it failed to identify most kids on the school”. That’s fair, we never even considered testing it with kids.