Google Blogoscoped

Friday, December 12, 2008

Mutating Pictures Video & Faces of Facemaker

I just uploaded a video made a while ago showing the progress of the Mutating Pictures experiment over several generations. To recap, in the experiment a pool of 1000 random symmetric polygon images was created, and everyone could then vote on how much an individual image looked like a face – with high-rated polygon mixes producing a lot of (mutated) offspring, and lower-rated producing little or no offspring. This way, generation by generation, the pool spread its more successful images and killed off the ones less face-like.

It’s interesting to note that the speed of the evolution in this experiment is not dependent on time, but rather on the number of people participating. In the first couple of days when many people participated, picture pool generations passed by quickly leading to many mutations and a fast face evolution (often the oldest living member in the pool of 1000 pictures was mere minutes old). After the first couple of days, traffic nearly died down, and so evolution came to a halt. The first days of the experiment created more power than the whole following year taken together:

Back during the experiment, Tomasz was inspired to put up a similar project with changed parameters. In his app Facemaker, instead of voting from 0 - 10, you simply picked which of two given images looked more face like. Tomasz also used rounder shapes. Still, it’s interesting to see how after time passed, the results of Mutating Pictures and Facemaker often look similar, despite different code bases and approaches. Here are some of Facemaker’s results:

[Video song is Erase by Control Escape, via YouTube’s AudioSwap feature.]


Blog  |  Forum     more >> Archive | Feed | Google's blogs | About


This site unofficially covers Google™ and more with some rights reserved. Join our forum!