The goal of this project is to use different deep learning techniques in order to merge the style of one image and the content of another image (neural style transfer).

The different techniques used are :

• A Neural Algorithm of Artistic Style by Gatys et al. [paper] : located in python/gatys/
• Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks by Zhu et al. [paper] : located in python/cycle-gan/

The different techniques have been implemented in python (version 3) mainly via the PyTorch library.

To create the necessary anaconda virtual environment (named neural_style_transfer), do

make create-env

or create the environment manually via the environment.yml file provided.

To use a technique, make sure to activate the virtual environment and do

python python/<technique-name>/main.py

Each hyperparameter of a technique can be modified via the corresponding main.py file.

This project was realized as part of the Deep learning course given by Professor Louppe to the master students of Civil Engineering (Computer Science) at the University of Liège during the academic year 2019-2020.

• Maxime Meurisse - meurissemax
• Adrien Schoffeniels - schoffeniels
• Valentin Vermeylen - ValentinVermeylen

• This project received a score of 17/20.