⚠️ This project is still in development. Please report bug but don't expect everything to work very smoothly.

A python module to streamline working with 2D and 3D vector graphics in machine learning projects.

Much of the philosophy of this module is based on Processing and p5.js. In other words it favors more accessible syntax over complete efficiency.

Single Image Notebook: ./notebooks/compose_test_01_py.ipynb

Video Notebook (very hacky): ./notebooks/compose_test_04_py.ipynb

Recommended python version is 3.7. Ubuntu based Linux is currently the only supported platform. If you run into issues on other platforms feel free to file an issue but it will be low priority.

conda create -n compose python=3.7 conda activate compose conda install -c conda-forge jupyterlab conda install -c anaconda ipykernel python -m ipykernel install --user --name=compose

sudo ./scripts/install/platform/ubuntu.sh sudo apt upgrade git submodule update --init --recursive python scripts/install/pytorch.py # to make sure you get the correct nvidia compile pip install -r requirements.txt cd diffvg python setup.py install cd .. python setup.py install

Currently documentation could use some work and can only be built locally. See CONTRIBUTING.md for some instructions.

This project would not be possible without the work of the creators of diffvg Tzu-Mao Li, Michal Lukáč, Michaël Gharbi, and Jonathan Ragan-Kelley.

Thanks to Peter Schaldenbrand, Zhixuan Liu, Jean Oh (creators of StyleCLIPDraw) for example of getting diffvg working on colab.