This Machine Can Tell You If Your Photos Suck Or RuleS

Xerox research labs have developed a computer program that tells you if your pictures are good or bad. In fact, it's capable of detecting if the photo has that magical, intangible quality that makes an image special. It works.

At least looking at their examples, which were obtained using an alpha version of their software, it seems it works great.

You can easily see the difference. The images on the left side are beautiful, strikingly pretty or dramatic. They have that something. They attract your eye instantly. The images of the right side are not that good. It's not that they are poorly made or blurry or have extremely bad compositions. They are... ok. They are flat, just one snoozer, blah and yawner after the other.

That's exactly what Xerox wanted to do, according to their labs people, their Aesthetic Image Search program is "trying to learn what makes an image special, and makes photo enthusiasts mark it as high quality."

Computer critic

Their critic software doesn't work with any photos. It is specialized in different subjects, like beaches, portraits, skies or flowers. The algorithm uses different parameters to evaluate the photos according to that subject matter, drawing conclusions that seem quite accurate most of the time.

This Machine Can Tell You If Your Photos Suck Or RuleS

I'm sure that some people will not agree with some of the judgements. Beauty is in the eye of the beholder and all that. But it's true that the Xerox critic art makes an excellent job at picking the remarkable stuff against the blah stuff.

I wonder what would happen if you feed it some apparently mundane photo like this Cindy Sherman, which was sold for $3.89 million. Or this bland Andreas Gursky's landscape, sold for $4.3 million. Would the algorithm say they are awesome or that they suck? Would it love Ansel Adams and Annie Leibovitz? Would Henri Cartier-Bresson confuse it or would it recognize the beauty of his work?

This Machine Can Tell You If Your Photos Suck Or RuleS

Perhaps the better question could be another: if you give this software a few rolls of Cartier-Bresson's photos, would it be able to distinguish what are the amazing ones, the ones that Henri picked himself in his lab, from the bad ones, the ones that were never published or printed?

Maybe if it ever achieves that we would be closer to create machines with a soul, computers that are not only smart but could appreciate beauty in the same way humans do. [Open XeroxThanks Oscar!]