/GANs_Artwork

Exploring the random generation of artwork via the usage of GANs

Primary LanguageJupyter NotebookMIT LicenseMIT

Creating Artwork with GAN's

This repository is meant to explore the random generation of artwork via the usage of GANs

Requirements

Run the following:

conda create -n art_gans python=3.7
pip install -r requirements.txt

Data

The abstract folder contains the data. This is a representation of where the data exists on your local.

root
├───data
|   ├───Abstract_GIF
|   ├───Abstract_Progression_Images
│   ├───cleaned_train
│   │   └───genre_of_art
│   ├───cleaned_validation
│   │   └───genre_of_art
│   ├───train
│   │   └───genre_of_art
│   └───validation
│       └───genre_of_art
├───model
│   └───model_name
└───util

Notebook

StyleGan2-ADA_Custom_Edited.ipynb is the notebook where the work resides. This notebook is google colab oriented. It seems that you need more than 12GB of RAM that is allocated for the free version unfortunately.

Util folder

Keeps all the preprocessing scripts for images

Executables

  • clean_data.py converts and resizes specifically JPG images to PNG.
    • Can convert JPG's to PNG's
    • Padd images to square shape and resize if appropriate
    • Check number of channels to see if there is an inappropriate shape
    • Convert images to GIF