Namgyu Kang · Jaemin Oh · Youngjoon Hong · Eunbyung Park

git clone https://github.com/NamGyuKang/Physics-Informed-Gaussians.git
cd Physics-Informed-Gaussians

Please follow the steps in the Jax_gpu_version_installation.txt file to install the JAX GPU version.

The code is tested with Python (3.8, 3.9) and PyTorch (1.11, 11.2) with CUDA (>=11.3). You can create an Anaconda environment with those requirements by running:

• if you use CUDA 11.3, Pytorch 1.11, Python 3.9, conda env create -f CUDA_11_3_Pytorch_1_11_Py_3_9.yml
• or with CUDA 11.6, Pytorch 1.12, Python 3.8, conda env create -f CUDA_11_6_Pytorch_1_12_Py_3_8.yml
• and then conda activate pig

• CUDA_VISIBLE_DEVICES=0 bash flow_mixing3d_pig.sh
• CUDA_VISIBLE_DEVICES=0 bash helmholtz2d_pig.sh
• CUDA_VISIBLE_DEVICES=0 bash klein_gordon3d_pig.sh
• CUDA_VISIBLE_DEVICES=0 bash diffusion3d_pig.sh

If you find this code useful in your research, please consider citing us!

@inproceedings{kang2025pig,
  title={{PIG}: {P}hysics-{I}nformed {G}aussians as {A}daptive {P}arametric {M}esh {R}epresentations},
  author={Namgyu Kang and Jaemin Oh and Youngjoon Hong and Eunbyung Park},
  booktitle={The Thirteenth International Conference on Learning Representations},
  year={2025},
  url={https://openreview.net/forum?id=y5B0ca4mjt}
}

Contact Namgyu Kang if you have any further questions.

This project is built on top of several outstanding repositories: SPINN, PIXEL, JAXPI. We thank the original authors for open-sourcing their excellent work.