Generating Paintings using Generative Adversarial Networks

An exploration into the research, design and implementation of bespoke Generative Adversarial Network (GAN) architectures is conducted, with the aim of synthesizing novel paintings that have ’aesthetic arousal’ and can be perceived as ’Art’ by observers.

View Paper

This is the repository for my Generation of Paintings using Generative Adversarial Networks Project at Imperial College London

The repository has been designed such that experiments for each architecture are self contained and are easily adaptable for any data-set. Each folder contains its own set of instructions in their respective README.md files to run experiments. It is important to note that many experiments run on different environments, some require Python 2.7, others require Python 3.6, as well as different versions of CUDA and cuDNN for compatibility reasons with older packages such as Torch. Thus, it is important to modify your environments accordingly to ensure certain experiments can be run. The repository contains nine folders:

• The ’Assets’ folder contains a small sample of generated images and gifs.
• The ’Clean Dataset’ folder contains code to clean the Xart data-set, which is necessary to run so that certain images that are corrupted and cannot be read by OpenCV Python package are removed.
• The ’DCGAN’, ’CDCGAN’, ’CAN’ and ’StackGAN’ folders contain the code for experiments run on these respective architectures.
• The ’Embeddings’ folder contains two folders ’CHAR-CNN-RNN’ and ’SkipThought’ that each contain the code for experiments conducted on these methods. Trained models from these methods can then be used in the StackGAN implementation.
• The ’Inception Score’ and ’FID Score’ folders contain the code to run these quantitative measures.
• It is important to note that some files may take a long time to train and test, sometimes over 3 weeks, thus saving weights and models every few epochs is crucial.

Datasets

The ArtImages data-set can be downloaded from this link

The WikiArt data-set can be downloaded from this link

*INSTRUCTIONS STILL NEED TO BE ADDED TO VARIOUS FOLDERS (OR YOU CAN FIGURE IT OUT) AND CERTAIN MODIFICATIONS NEED TO BE UPLOADED