ProofByConstruction / texture-networks

Implementation of Texture Networks in Tensorflow
27 stars 9 forks source link

Texture Networks in Tensorflow.

The short-term goal of this project is to implement Texture Networks, as described in http://arxiv.org/abs/1603.03417. (For more background, see http://arxiv.org/abs/1508.06576) Longer term goals include exploring extensions and related architectures around the theme.

Development is being done in python 3.5, though you will need to use 2.7 to run Tensorboard.

Setup

You will need to get a copy of the VGG16 network, I recommend the torrent listed at https://github.com/ry/tensorflow-vgg16

Set up a python 3.5 virtual environment, intall tensorflow and then pip install -r requirements.txt. (If that scares you, I think the only real reqs so far are tensorflow, scikit, and pillow. If you hit problems let me know.) You should then be good to run python texture_network.py and python vgg_network.py (after pointing the latter at your local vgg file, and adding an img.jpg to the img directory).