Monet Moves

Andy Cai, Stephanie Doan, Michael Inoue, Henry Trinh

Overview

We use three models for the tasks of art style transfer from Claude Monet's paintings to real photographs and artwork generation from existing artwork:

1. CycleGAN based on Pytorch implementation
2. Neural style transfer featuring perceptual loss for art style transfer, based on Keras implementation
3. DCGAN based on Pytorch implmentation

Setup

Dataset

Download Kaggle dataset of Monet's paintings and real photographs (png files) and put into /data folder with the following file structure.

monet-moves
  data
    monet
      monet_
        0a5076d42a.jpg
        0bd913dbc7.jpg
        ...
    photo
      0a0c3a6d07.jpg
      0a0d3e6ea7.jpg
      ...

Virtual Environment

Install and run virtualenv, install dependencies.

pip3 install virtualenv
python3 -m virtualenv env
source env/bin/activate
pip3 install -r requirements.txt

Save dependencies and deactivate.

pip3 freeze > requirements.txt
deactivate