Deepdream prototypes [2015]

A series of images produced using Google’s inceptionism/deepdream code released in July 2015. The software uses artificial neural networks trained on large sets of classified images used by Google for image recognition. The code accepts images as inputs and iteratively evolves the pixel values towards some coherent resemblance to the image classes it knows, producing wild images of “pig-snails,” “camel-birds,” “dog-fish,” and the famous “puppy slugs,” among many other categories.

I was able to get interesting animations by iteratively zooming into the output and oscillating which of the layers to enhance in the network.

Enhancing different layers (like 'inception_3a/relu_5x5_reduce') let me find some of the lesser-seen classes like these chalice-like bowls.

A gallery of images including paintings by Da Vinci, Klimt, Monet, Schiele, and Picasso, photos of Bagan, Lake Indawgyi, and Shwedagon Pagoda in Burma/Myanmar, Jaaga in Bangalore, and some fireworks.