https://github.com/milosgajdos/creep-dreamz

Deep Dream experiments inspired by Keras Deep Dream
https://github.com/milosgajdos/creep-dreamz

deep-dream deep-dream-experiments deeplearning keras-deep-dream neural-networks

Last synced: about 2 months ago
JSON representation

Deep Dream experiments inspired by Keras Deep Dream

• Host: GitHub
• URL: https://github.com/milosgajdos/creep-dreamz
• Owner: milosgajdos
• License: apache-2.0
• Created: 2018-02-18T23:32:14.000Z (over 7 years ago)
• Default Branch: master
• Last Pushed: 2022-11-22T04:28:32.000Z (over 2 years ago)
• Last Synced: 2025-04-06T10:02:44.335Z (about 2 months ago)
• Topics: deep-dream, deep-dream-experiments, deeplearning, keras-deep-dream, neural-networks
• Language: Python
• Homepage:
• Size: 991 KB
• Stars: 1
• Watchers: 0
• Forks: 1
• Open Issues: 7
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# creep-dreamz

Deep Dream experiments

This project is a "cleaned up" rewrite of [keras Deep Dream](https://github.com/keras-team/keras/blob/master/examples/deep_dream.py). 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](https://www.tensorflow.org/api_guides/python/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](https://commons.wikimedia.org/wiki/File:Handsome-man-by-Willy-Volk-Creative-Commons.jpg)