sparse-conv-gpis-shadertoy - GitLite

KO | EN

gitlite — search

#terminal #golang #tensorflow #bash #generative-ai #ai-agents #ai-agent #php #java #nextjs

sparse-conv-gpis-shadertoy

(★ 10)

Source code (Shadertoy) for the SIGGRAPH Asia 2025 paper "Practical Gaussian Process Implicit Surfaces with Sparse Convolutions".

파일 탐색기

ZIP 다운로드

- - Uffizi_down.jpg
  - Uffizi_east.jpg
  - uffizi_latlong.jpeg
  - Uffizi_north.jpg
  - Uffizi_south.jpg
  - Uffizi_up.jpg
  - Uffizi_west.jpg
- - negx.jpg
  - negy.jpg
  - negz.jpg
  - posx.jpg
  - posy.jpg
  - posz.jpg
  - readme.txt
- shadertoy_2d_gpis.png
- shadertoy_3d_gpis.png
- shadertoy_ndf_visualization.png
- shadertoy_weight_space.png
- knob.glsl
.gitignore
accumulate_gpis_2d_microfacet.glsl
accumulate_gpis_3d.glsl
common.glsl
gaussian_process.glsl
gpis_2d_microfacet.glsl
gpis_3d.glsl
gui.glsl
hash.glsl
LICENSE
ndf_visualization.glsl
README.md
weight_space.glsl

# CDN으로 사용하기

jsDelivr

jsDelivr는 공개 GitHub 리포지토리를 별도 설정 없이 CDN으로 즉시 서빙합니다. 버전과 파일을 고르면 웹페이지에 바로 붙일 수 있는 링크와 예시 코드가 만들어집니다.

버전

파일

링크

예시

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

### [Project Page](https://cs.dartmouth.edu/~wjarosz/publications/xu25practical.html) | [Paper](https://cs.dartmouth.edu/~wjarosz/publications/xu25practical.pdf) | [C++ Code](https://github.com/dartmouth-vcl/sparse-conv-gpis-tungsten)

This repository contains the Shadertoy code for the SIGGRAPH Asia 2025 paper:

> **Practical Gaussian Process Implicit Surfaces with Sparse Convolutions**<br>
> Kehan Xu (Dartmouth College), Benedikt Bitterli (NVIDIA), Eugene d'Eon (NVIDIA), Wojciech Jarosz (Dartmouth College)

The code implements a simplified light transport model and serves as a proof of concept showing that our sparse convolution GPIS representation is GPU-friendly. A version supporting full light transport on CPU is implemented based on [Tungsten](https://github.com/tunabrain/tungsten); please refer to [this repository](https://github.com/dartmouth-vcl/sparse-conv-gpis-tungsten).

# Overview
You can run the Shadertoy demos locally using the ShaderToy extension in VS Code, or view them online on [Shadertoy.com](https://www.shadertoy.com/). This repository includes four demos, each with detailed documentation provided through in-file comments.

The supplementary video on the project page showcases the demos included in this repository.

## 2D GPIS and Microfacet ([Online](https://www.shadertoy.com/view/33BBDh))

`accumulate_gpis_2d_micorfacet.glsl`

<img src="imgs/shadertoy_2d_gpis.png" alt="2D GPIS" width="1024"/>

The core implementation resides in `gpis_2d_micorfacet.glsl`. `accumulate_gpis_2d_micorfacet.glsl` serves as a wrapper to accumulate samples when the camera remains static. This demo shows a single realization and the ensemble average of a 2D GPIS, and demonstrates how to match the appearance of the GPIS ensemble average with the microfacet BRDF with a Beckmann NDF.

## 3D GPIS ([Online](https://www.shadertoy.com/view/W3BBDh))

`accumulate_gpis_3d.glsl`

<img src="imgs/shadertoy_3d_gpis.png" alt="3D GPIS" width="1024"/>

The core implementation resides in `gpis_3d.glsl`. This demo shows a single realization and the ensemble average of a 3D GPIS. It also allows toggling between sampling in world space and isotropic space, as well as visualizing the surface normal.

## NDF Visualization ([Online](https://www.shadertoy.com/view/3XBBDh))

`ndf_visualization.glsl`

<img src="imgs/shadertoy_ndf_visualization.png" alt="NDF Visualization" width="1024"/>

This demo visualizes the normal distribution function (NDF) of a GPIS on a hemisphere, both through sampling and analytic derivation, and shows that the two agree closely.

## Weight Space vs. Sparse Convolution ([Online](https://www.shadertoy.com/view/WXBBDh))

`weight_space.glsl`

<img src="imgs/shadertoy_weight_space.png" alt="Weight Space vs Sparse Convolution" width="1024"/>

This demo shows that both the weight-space realization and the sparse convolution realization can represent the same stationary GPIS, while the sparse convolution approach can additionally represent non-stationary GPISes.

## License and Acknowledgements

Portions of this code are adapted from other projects. We acknowledge and thank the original authors for their contributions:

- [A Dataset and Explorer for 3D Signed Distance Functions](https://github.com/tovacinni/sdf-explorer) — by **Towaki Takikawa**  
- [ShaderToy: "Distance field outline mask"](https://www.shadertoy.com/view/XtG3Rt) — by **Good**  
- [ShaderToy: "Simple Bitmap Text "](https://www.shadertoy.com/view/4dtGD2) — by **Hamneggs**  
- [ShaderToy: "Hash without Sine"](https://www.shadertoy.com/view/4djSRW) — by **David Hoskins**
- [ShaderToy: "Hash Functions for GPU Rendering"](https://www.shadertoy.com/view/XlGcRh) — by **Mark Jarzynski and Marc Olano**

All newly written code in this repository is provided under the **BSD 3-Clause License**, the text of which can be found in [`LICENSE`](./LICENSE).

# Citation
If you find this code useful in your academic research, please cite the original paper:
```
@article{xu25practical,
    author  = {Xu, Kehan and Bitterli, Benedikt and d'Eon, Eugene and Jarosz, Wojciech},
    title   = {Practical {G}aussian Process Implicit Surfaces with Sparse Convolutions},
    journal = {ACM Transactions on Graphics (Proceedings of SIGGRAPH Asia)},
    year    = {2025},
    month   = dec,
    volume  = {44},
    number  = {6},
    doi     = {10.1145/3763329}
}
```