Artistic style transfer implementation with Tensorflow

This project is intended to be the course final project for Columbia University GR5242 Advanced Machine Learning.

All copyrights reserved by author: Tian Gao (CU UNI: tg2585)

Directory Structure

project/
   |---- input/
   |       |---- content/
   |       |        |---- content.jpg
   |       |---- style/
   |       |        |---- style.jpg
   |---- output/
   |       |---- output.jpg
   |
   |---- style_transfer.py
   |---- neural_network.py
   |---- visual_geometry_group.py
   |
   |---- utils.py
   |---- constants.py
   |---- settings.py
   |---- logger.py
   |
   |---- imagenet-vgg-verydeep-19.mat
   |---- requirements.txt

This directory structure only shows directories and files that are necessary to run the code and generate certain outputs.

This repository also includes a report folder which contains the full project report (source files, illustrations and PDF), which is exclusively for Columbia University GR5242. All copyrights reserved.

Running

1. Construct your folder exactly as of the structure above; put the content picture (the one to be transfered) in content folder and put the style picture (the one that provides with the artistic style) in style folder

2. Make sure you have the pre-trained neural network data from VGG

3. This project runs with Python 3.6. Install dependencies with

   pip install -r requirements.txt
   

   and then run the following command

   python style_transfer.py --content content.jpg --style style.jpg --output output.jpg
   

   Change *.jpg in the command input to match your file names

Related Works

This project is inspired by the paper on style transfer.

Acknowledgement

Thanks to the authors of the paper, who had such interesting ideas and inspired this work.