Besserwisser post on EvoLisa
Posted by Dan Byström on December 14, 2008
In short: would it be possible to construct a vector version of some image by overlaying just a few semi-transparent polygons? And if so, how do you figure out how these polygons would look like?
I can’t figure out how he got this idea. If someone had asked me if it could be done I’m afraid I’d just answered “Heck, no. No way. No use even trying.”. Therefore; a brilliant idea I must say!
Some people saw it as a proof of evolution. Some people saw it as proof of creationism. Some people saw it as an image compression algorithm. Some people saw it as a cool but useless toy. I guess some other people didn’t know what to think.
I guess I saw it as a powerful demonstration of a technique to use when you really have no idea how to attack a really complicated problem. I wounder if someone know how to construct an algorithm that directly converges towards the desired image without using randomness? The word “tricky” really sounds like an understatement!
Out of curiosity, I looked at the fitness function. That is, the piece of code that tries to compare how similar/different two images are. Then I noticed that since Roger apparently had been under much pressure to reveal his quick and dirty hack to the public, he hadn’t had the time to optimize that code. So I couldn’t resist doing just that.
And behold: the performance increase was a whopping 25 times! Yeah,a 25 time speed increase is like cruising in 50 km/h compared to breaking the sound barrier. Pretty cool, eh?