Deep Dream

Minimal Python/TensorFlow implementation of the DeepDream algorithm originally created by Alexander Mordvintsev.

Usage

To use on the predefined .jpg image:

python3 deep_dream.py <path_to_the_image>

or

python3 deep_dream.py

to perform on the random image.

How does it work?

We are using Inception5h model which was designed to classify images.

During the classification process we are providing input images and using gradient descent to adapt weights to the images through filters.

DeepDream algorithm does the opposite. It adapts the input images to match the network weights with gradient ascent which results in visualizing network filters on the input images giving them psychodelic look.

Results

Author

Greg (Grzegorz) Surma

PORTFOLIO

GITHUB

BLOG