This GitHub repository is an extension of the original GitHub repository https://github.com/google/deepdream from Alexander Mordvintsev, Michael Tyka and Christopher Olah from Google. The original Readme is at the end of this file; the same licence(file LICENCE) applies for this work.
The Installation of DeepDream is quite delicate; see this tutorial http://www.knight-of-pi.org/installing-the-google-deepdream-software/ for installing DeepDream on a Ubuntu 14.04 machine and this tutorial http://www.knight-of-pi.org/deepdreaming-on-a-raspberry-pi-2/ for installing DeepDream on a Raspberry Pi 2.
The script deepdreaming.py is controlled via command line parameters. They are listed with
$python deepdreaming.py --help
The source file is chosen with the parameter -s:
$python deepdreaming.py -s sky_1024.jpg
Select the number of iterations with the parameter -i:
$python deepdreaming.py -s sky_1024.jpg -i 3
Be carefull, though, since choosing a large number here freezes the computer for quite some time.
The depth of the dream could be chosen with the -d flag and a number between 1 and 10:
$python deepdreaming.py -s sky_1024.jpg -d 3
The layer type in which the dream should end could be chosen with the -t flag and a number between 1 and 6:
$python deepdreaming.py -s sky_1024.jpg -t 1
Choose an other octave(index of ranked prediction results, i guess) with the -o flag:
$python deepdreaming.py -s sky_1024.jpg -o 8
A guide file for a guided Dream could be attached with the -g flag. This works best with a reduced depth, e.g. -d 2:
$python deepdreaming.py -s sky_1024.jpg -g flowers.jpg -d 2
Set the -r flag to choose octave, layer depth and type randomly:
$python deepdreaming.py -s sky_1024.jpg -r
The output filename will be the higest 5-digit integer number in ./dreams which is not occupied already, e.g. "00023.jpg".
This repository contains IPython Notebook with sample code, complementing Google Research blog post about Neural Network art. See original gallery for more examples.
You can view "dream.ipynb" directly on github, or clone the repository, install dependencies listed in the notebook and play with code locally.
It'll be interesting to see what imagery people are able to generate using the described technique. If you post images to Google+, Facebook, or Twitter, be sure to tag them with #deepdream so other researchers can check them out too.