GSCache: Real-Time Radiance Caching for Volume Path Tracing using 3D Gaussian Splatting

David Bauer, Qi Wu, Hamid Gadirov, Kwan-Liu Ma
IEEE VIS 2025
Project Page / Paper

Official implementation of the 2025 IEEE VIS paper "GSCache: Real-Time Radiance Caching for Volume Path Tracing using 3D Gaussian Splatting"

Build Instructions

The build was tested on Ubuntu 22.04 with CUDA 12.8. Windows and Mac might work but are not tested.

Dependencies

These are the dependencies the project was tested with. Earlier versions might work.

• CMake 3.20+
• C++17 Compiler
• CUDA 12.5+
• LibTorch 2.7+

Building

First edit the file setup_env.sh to correctly point to your CUDA, LibTorch installation and GPU architecture. Use the build script to build the project:

On Linux:

    source setup_env.sh
    sh build.sh

Updates

• ✅ [2025/12] Open source release
• ✅ [2025/11] Conference presentation in Vienna
• ✅ [2025/07] Paper accepted to IEEE VIS 2025

BibTex

@misc{bauer2025gscache,
      title={GSCache: Real-Time Radiance Caching for Volume Path Tracing using 3D Gaussian Splatting}, 
      author={David Bauer and Qi Wu and Hamid Gadirov and Kwan-Liu Ma},
      year={2025},
      eprint={2507.19718},
      archivePrefix={arXiv},
      primaryClass={cs.GR},
      url={https://arxiv.org/abs/2507.19718}, 
}

License

The source code for this project is open under the Apache 2.0 license. 3D Gaussian splatting code and third party libraries are subject to their own licenses as noted in file headers and license files.