Skin tones in images are very sensitive to lighting conditions. How are they planning to normalize these differences?

In HSV space it's not as much of a problem. People fit into a black-white-red triangle

I don't understand why the choice of colorspace used to represent the person's face matters. The lighting used when the photo was taken can alter the apparent skin tone, possibly to another natural looking skin tone.

Because you are using relative colors so the brightness ()and to an extent the color) of lighting largely cancels out.

Unless of course you take a very pale blonde person and light them with a brown spotlight!

Could you please explain this in more depth, or provide a link?

Saw this a while ago and it's a cool thought experiment. The application fails, though, as some HNers have pointed out. It looks like Time covers have simply become darker, independent of the skin tones on each cover.

Someone points out the lack of "image normalization" in his dataset. I'm surprised he didn't take this first step, but this looks like more of a weekend/toy project than something extremely rigorous.

Lighting conditions make this impractical. I clicked on several of the darker brown colors and they all had white people of European origin. In fact, after clicking a while on dark brown I could only find white people!

I clicked for a while longer, still only european looking or caucasian people. The only candidate I clicked on that actually was naturally brown was the picture of a chocolate (september 12th 2012). I'm going to stop clicking now, this site is a waste of my time as the underlying technique is proven to be a fiasco.