voxgan 0.0.2

Python package to generate 2D and 3D images using Generative Adversarial Networks and PyTorch

voxgan

voxgan is a Python package to generate 2D and 3D images using Generative Adversarial Networks and PyTorch. It was developed to help along the project FluvGAN, with no specific development planned beyond this project.

Installation

You can directly install voxgan from pip:

pip install voxgan

Or from GitHub using pip:

pip install git+https://github.com/grongier/voxgan.git

To run the Jupyter notebook in examples you will also need pandas, matplotlib, and jupyter, which you can install from pip too:

pip install pandas matplotlib jupyter

Usage

Here's a basic example of voxgan's API:

from voxgan.models.dcgan import DCGAN3d

# Define a PyTorch dataset for training
training_dataset = # TO DEFINE

# Define a GAN based on DCGAN, which includes generator and discriminator
gan = DCGAN3d()
# Configure the GAN for training, including the optimizers
gan.configure()
# Train the GAN
gan.train(training_dataset)
# Generate 10 samples
samples = gan.predict(num_samples=10)

For a more complete example, see the Jupyter notebook using_voxgan.ipynb in examples. For more advanced uses of voxgan, see FluvGAN.

Citation

If you use voxgan in your research, please cite the original article(s) describing the method(s) you used (see the docstrings for the references). You can also cite voxgan itself:

Rongier, G. (2021) voxgan. v1. CSIRO. Software Collection. https://doi.org/10.25919/cdgf-cw44

Here is the corresponding BibTex entry if you use LaTex:

@misc{Rongier2021,
    author = "Rongier, Guillaume",
    title = "voxgan",
    number = "v1",
    institution = "CSIRO",
    type = "Software Collection",
    year = "2021",
    doi = "10.25919/cdgf-cw44"
}

Credits

This software was written by:

Guillaume Rongier

License

Copyright notice: Technische Universiteit Delft hereby disclaims all copyright interest in the program voxgan written by the Author(s). Prof.dr.ir. S.G.J. Aarninkhof, Dean of the Faculty of Civil Engineering and Geosciences

© 2021-2025, Guillaume Rongier
© 2020-2021, Commonwealth Scientific and Industrial Research Organisation (CSIRO) ABN 41 687 119 230

This work is licensed under a MIT OSS licence, see LICENSE for more information.