Google’s Deep Dream neural network tends to hallucinate in dog faces, because the database it trained on had a disproportionately high number of dogs pictures in it. But what if that database had been full of cats instead? You’d end up with something like this: a neural network that can dream about an infinite number of imaginary felines.
The Cat Generator is a coding experiment by Alexander Jung, which uses neural network analysis on a data image set to randomly generate images of individual cats from the ditital ether. The resulting cats look a little weird close-up, but as thumbnails, these all look like real kitties of varying breeds, color, and manginess.
An excellent start. Of course, the ultimate goal for a project like this should be to give the computer access to the Impact font and randomly generate LOLCats forever. Make it happen, Mr. Jung!
[via Prosthetic Knowledge]