Today, a friend attended me of a new(?) research project, which tries to rate the aesthetic appeal of pictures by using artificial intelligence.
While I am fascinated by the idea, it seems that this still has a long way to go, and that the designers maybe focus too much on heuristics (people like a small DoF, or vanishing points, or strong colors), and less on the human context and background when evaluating a photograph.
I don't know if there is something like universal aesthetics. Doesn't cultural background heavily influence our likes and dislikes? Besides, would it be really possible to catch the appeal of a picture in a set of rules that can be evaluated by a computer?
However, it is fun to throw one's pictures at Acquine, and try to find out what it likes and doesn't like. This one gets 85%, for instance, while this one only gets 6%. Funny enough, the latter picture does much more for me, because of my personal interests. But I think Acquine doesn't see enough clear lines or division between subject and background.
Both are my own pictures, btw ... no "stealing" of other's work ;-)
I thought that a well-structured background with human characters in the foreground would get a good score, and I tried this one:
which was rated 89.1
I browsed photos by other users. I found one with a very high score, which was a completely missed photo: a door opening on a room with a character inside. The human character was underexposed, the door frame was overexposed.
Another one with a crossed-dressed drunkard in a party, badly exposed with a point-and-shoot and a flash got an incredible 95.9 (http://acquine.alipr.com/show_image.php?id=1255569777.6365).
Most images are not bad, but I think the software measures only some sort of 'eye-catchingness'.
Johan has replied to François CollardFrançois Collard has replied to JohanMaybe you could create a group for images rated > 80?
Sign-in to write a comment.