glint24

KO | EN

gitlite — search

#javascript #python #hacktoberfest #react #ai #typescript #llm #go #golang #android #machine-learning #rust #deep-learning #linux

★ 9 Open GitHub ↗

Position-Normal Manifold for Efficient Glint Rendering on High-Resolution Normal Maps

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Position-Normal Manifold for Efficient Glint Rendering on High-Resolution Normal Maps
This repo contains the implementation of the SIGGRAPH 2025 paper: **Position-Normal Manifold for Efficient Glint Rendering on High-Resolution Normal Maps**.

- `src/bsdfs/` contains the Mitsuba 0.6 implementation of specular glint BRDF and diffuse BRDF.

- `convert.py` converts input normal map to the cluster and bounding box hierarchy.

- `convert_G.py` converts input normal map into GGX roughness and tangent frame for fast shadow-masking approximation.

### [Paper](https://arxiv.org/abs/2505.08985) | [Citation](#citation)

## Setup
- Mitsuba 0.6
- CUDA 11.7
- Python 3.8
- pytorch 2.0.8
- opencv-python 4.8.0.76

## Usage

BSDF plugin in Mitsuba 0.6's XML format:

`````` xml
<bsdf type="glintconductor">
            <string name="micro_normal" value=$normalmap_folder/>
            <float name="sigma" value=$footprint_scale/>
            <float name="thresh" value="0.0004"/> 
            <boolean name="fix_sigma" value="false"/> 
            <float name="texture_scale" value="16.0"/> 
            <integer name="res" value="1024"/> 
            <string name="filter" value="gaussian"/> 
</bsdf>

<bsdf type="glintdiffuse">
            <rgb value="0.8 0.8 0.8" name="reflectance"/>
            <string name="micro_normal" value=$normalmap_folder/>
            <float name="sigma" value=$footprint_scale/>
            <boolean name="fix_sigma" value="false"/> 
            <float name="texture_scale" value="16.0"/>
            <integer name="res" value="1024"/>
</bsdf>
``````

Normal map to cluster and bounding box hierarchy:

```shell
python convert.py --input <normal-or-height-map-file> --output <output-folder>
```

An example is provided in `demo/` that reproduces Fig. 11 of the paper:
1. Download [Mitsuba 0.6](https://github.com/mitsuba-renderer/mitsuba).
2. Copy `src/bsdfs/` to the Mitsuba 0.6 folder and add lines below to `src/bsdfs/SConscript`
```python
plugins += env.SharedLibrary('glintdiffuse', ['glintdiffuse.cpp'])
plugins += env.SharedLibrary('glintconductor', ['glintconductor.cpp'])
```
3. Compile according to the [documentation](https://mitsuba-renderer.org/docs.html).
4. In this folder, `cd demo` and run `sh demo.sh`.
5. The rendering results will be saved as `scene1_isotropic.exr`, `scene1_brush.exr`, `scene1_scratched.exr`.

## Citation

``````
@inproceedings{wu2025position,
	author = {Liwen Wu and Fujun Luan and Miloš Hašan and Ravi Ramamoorthi},
	title = {Position-Normal Manifold for Efficient Glint Rendering on High-Resolution Normal Map},
	booktitle = {SIGGRAPH},
	year = {2025}
}
``````

glint24

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

glint24

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README