README.md

creep-dreamz

Deep Dream experiments

This project is a "cleaned up" rewrite of keras Deep Dream. It provides a CreepDream class which implements Deep Dream based on the supplied parameters. The project no longer depends on scipy module for image manipulation that was replaced by TensorFlow native tf.image image manipulation module.

Example Usage

You can find a simple example in main.py file which demonstrates how to use CreepDream:

usage: main.py [-h] -i INPUT -o OUTPUT -m MODEL [-iter ITERATIONS] [-s STEP]
               [-oct OCTAVE] [-ocs OCTAVESCALE] [-mxl MAXLOSS]

Creep Dreamz with Keras.

optional arguments:
  -h, --help            show this help message and exit
  -i INPUT, --input INPUT
                        Path to the input data
  -o OUTPUT, --output OUTPUT
                        Path to the output data
  -m MODEL, --model MODEL
                        Keras model name
  -iter ITERATIONS, --iterations ITERATIONS
                        Number of gradient ascent steps per scale
  -s STEP, --step STEP  Gradient ascent step size
  -oct OCTAVE, --octave OCTAVE
                        Number of scales at which to run gradient ascent
  -ocs OCTAVESCALE, --octavescale OCTAVESCALE
                        Size ratio between scales
  -mxl MAXLOSS, --maxloss MAXLOSS
                        Maximum gradient ascent loss

Example run results

python3 main.py -i "random_man.jpg" -o "random_man_creep.png" -m "InceptionV3"  -oct 4 -ocs 1.4 -s 0.01 -iter 15 -mxl 10.0

Original image:

Creep Dreamt image:

[1] Image source: https://commons.wikimedia.org/wiki/File:Handsome-man-by-Willy-Volk-Creative-Commons.jpg