A Deep Dream of a Neural Network
A deep dream with TensorFlow's inception model.
It's a deep dream through the inception v3 model's layers that's trained on imagenet, and then the dream is allowed to run wild through various layers of the network, generating things that represent the various layers. Think of it like letting the neural network "amplify" what it thinks it sees...we can then amplify this amplification continuously and maybe manipulate the image, by, say, zooming in a bit. For example, the earlier layers are more like straight lines, then squares, then more circles....then things like eyes, maybe some faces and buildings...etc.
A tutorial on how this was done to come eventually.
Music:
Blue Sizzle - Madness Paranoia by Kevin MacLeod is licensed under a Creative Commons Attribution license (https://creativecommons.org/licenses/by/4.0/)
Source: http://incompetech.com/music/royalty-free/index.html?isrc=USUAN1100385
Artist: http://incompetech.com/