R3GW: Relightable 3D Gaussians for Outdoor Scenes in the Wild

Project Page · arXiv

Installation and Dataset

Clone the repository:

git clone git@github.com:fraunhoferhhi/R3GW.git
cd R3GW

Create the environment:

conda env create --file environment.yml
conda activate R3GW

Install manually nvdiffrast:

pip install git+https://github.com/NVlabs/nvdiffrast.git --no-build-isolation

To train and evaluate our model we used the version of the NeRF-OSR dataset provided by LumiGauss, which can be downloaded here.

Our model requires sky masks for training. For each scene, these masks must be generated in advance and stored in a floder named sky_masks within the directory undistorted. We generated the sky masks from the segmentation masks provided by NeuSky.

Usage

The script run_all.sh contains instructions to train, render, and evaluate all scenes, while run_relighting.sh provides a relighting example.

Acknowledgments

We acknowledge the following amazing repositories that contributed to the development of our code:

• 3D Gaussian Splatting for Real-Time Radiance Field Rendering
• LumiGauss: Relightable Gaussian Splatting in the Wild
• GaussianShader: 3D Gaussian Splatting with Shading Functions for Reflective Surfaces
• nvdiffrec
• Spherical Harmonics by Andrew Chalmers

Citation

If you use our method in your research, please cite our paper. You can use the following BibTeX entry:

@InProceedings{corona2026r3gw,
  author    = {Margherita Lea Corona and Wieland Morgenstern and Peter Eisert and Anna Hilsmann},
  title     = {R3GW: Relightable 3D Gaussians for Outdoor Scenes in the Wild},
  booktitle = {Proceedings of the 21st International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISAPP 2026)},
  year      = {2026},
  pages     = {432-443},
  publisher = {SCITEPRESS - Science and Technology Publications},
  doi       = {https://doi.org/10.5220/0014332200004084}
}