This project has been going for a while, so I’ve decided to write a short overview of the results seen so far. If you don’t know what’s it about – The Living Image project is a website where anyone can vote for one or more digitally generated images. Those pictures are created by artificial neural networks that evolve – generation by generation – based on viewers votes.
It appears that images generated were becoming increasingly complex for a couple of generations, while now they seem to have stabilised. On the other hand, there are some quite interesting images in the eighth generation (like this eye thingy), so I could be wrong 😛
The user activity has been sporadic, taking 2-4 weeks to gather approximately 300 votes necessary to create another generation. So go on and vote! 🙂
Over time I’ve made some modifications to the image generation tool, removing recurrent links from neural networks, reducing the number of activation iterations and adding a “sub-sampling” algorithm to (hopefully) prevent noise-like artefacts from appearing in the images.
This new and revised source code is now avalailable for download : wslivingimage.zip (8 KB).
Be warned that this program was initially created for internal use, so the interface is unintuitive and the code is only sparsely commented. Still, it might be useful to look at as an example of DelphiNEAT library usage.
That’s it for now 🙂Related posts :