Ref-GS : Directional Factorization for 2D Gaussian Splatting

Project Page | Paper | arXiv

Ref-GS : Directional Factorization for 2D Gaussian Splatting
Youjia Zhang, Anpei Chen, Yumin Wan, Zikai Song, Junqing Yu, Yawei Luo, Wei Yang
CVPR 2025

⚙️ Setup

Install Environment via Anaconda (Recommended)

conda create -n ref_gs python=3.7.16
conda activate ref_gs
pip install -r requirements.txt

pip install submodules/diff-surfel-rasterization-real
pip install submodules/diff-surfel-rasterization
pip install submodules/diff-surfel-2dgs
pip install submodules/simple-knn

git clone https://github.com/NVlabs/nvdiffrast
pip install .

📦 Dataset

We mainly test our method on Shiny Blender Synthetic, Shiny Blender Real, Glossy Synthetic and NeRF Synthetic dataset. Please run the script nero2blender.py to convert the format of the Glossy Synthetic dataset.

Put them under the data folder:

data
└── refnerf
    └── car
    └── toaster
└── nerf_synthetic
    └── hotdog
    └── lego

🏃 Training

We provide the script to test our code on each scene of datasets. Just run:

sh train.sh

You may need to modify the path in train.sh

✍️ Test

We provide simple jupyter notebooks notebook/test.ipynb to explore the model.

For mesh extraction, we adopt the same method as used in 2DGS.

🫡 Acknowledgments

This work is built on many amazing research works and open-source projects,

• NeRF-Casting: Improved View-Dependent Appearance with Consistent Reflections
• 3D Gaussian Splatting with Deferred Reflection
• 2DGS: 2D Gaussian Splatting for Geometrically Accurate Radiance Fields

We are grateful to the authors for releasing their code.

📜 Citation

If you find our work useful in your research, please consider giving a star ⭐ and citing the following paper 📝.

@article{zhang2024ref,
  title={Ref-GS: Directional Factorization for 2D Gaussian Splatting},
  author={Zhang, Youjia and Chen, Anpei and Wan, Yumin and Song, Zikai and Yu, Junqing and Luo, Yawei and Yang, Wei},
  journal={arXiv preprint arXiv:2412.00905},
  year={2024}
}

✉️ Contact

For feedback, questions, or press inquiries please contact Youjia Zhang.