This summer, Google published an article on their research into training neural networks in recognizing features of an image, and what would eventually happen if you run the network “backwards”, allowing the trained net to “dream up” stuff from a given image.
We will write more on this highly interesting topic, but we are seeing that after the first flood of images featuring slug-and-dog-infested vacation pictures, usage is now shifting into what can best be described as “smart” image filters.
Here are some simple examples to show where these things can take us.
Eight variations of the first original image, showing a small selection of the transformations
Original photo is zoomed during transformation to add further variation
