Another reason why I wish for NUCLEAR HOLOCAUST, in 1 minute
http://news.yahoo.com/s/time/20100122/wl_time/08599195464300
This world is retarded
Another reason why I wish for NUCLEAR HOLOCAUST, in 1 minute
http://news.yahoo.com/s/time/20100122/wl_time/08599195464300
This world is retarded
Saw that about a month ago on youtube
I wonder whether they knew that the feature is lacking or if they just plain didn't test for that.
Originally posted by dadudemon
It's the latter.
You sure? It seems such an obvious thing to test for. I mean, if I engineered that and started testing after like two or three real, caucasian people I'd be like "Neat, I wonder if it works on them others, too". And then I'd pay a black person to stand in front of that, cause I don't have any black friends or co-workers.
Originally posted by Bardock42
You sure? It seems such an obvious thing to test for. I mean, if I engineered that and started testing after like two or three real, caucasian people I'd be like "Neat, I wonder if it works on them others, too". And then I'd pay a black person to stand in front of that, cause I don't have any black friends or co-workers.
You wouldn't pay a black person.
Originally posted by Bardock42
You sure? It seems such an obvious thing to test for. I mean, if I engineered that and started testing after like two or three real, caucasian people I'd be like "Neat, I wonder if it works on them others, too". And then I'd pay a black person to stand in front of that, cause I don't have any black friends or co-workers.
I'm positive. The idea that it may not work on a black person would not cross most people's minds as the software is picking up on shadows on the face, not the skin tone. There'd be no logical reason to check to see if it works on dark-skinned people, based on the nature of the program....as black people would have shadows, as well. The problem is due to the lack of contrast between the shadows and the surrounding skin, which is what the HP engineers overlooked.
I assume that the program algorithm just looks for a contrast differential based on specific shapes (common to the face). So, it will look for the cresecent shaped shadow under the eyebrow, and run a contrast comparison against the surrounding skin (say, it looks for a 80+ point difference (on a 255 point scale)).
It's also possible that the testing lab for this software had different lighting than they could account for, so they really did test it with darker skinned people. In that factory video, the camera is oddly adjusting light sensitivity, making the black dude look really dark. That could be part of the problem: maybe the algorithm works just fine...they just need to adjust the light sensitivity and how it adjusts?