Physically Inspired Gaussian Splatting for HDR Novel View Synthesis

CVPR 2026

Huimin Zeng, Yue Bai, Hailing Wang, Yun Fu

Northeastern University

Bathroom	Chair
Sofa	Diningroom
Dog	Bear

Overview

PhysHDR-GS models scene appearance via intrinsic reflectance and adjustable ambient illumination.

Key Features:

• ✅ Physically inspired dual-branch design (IE + GI branch)
• ✅ Explicit HDR self-supervision without ground truth
• ✅ Reducing exposure-biased gradient starvation
• ✅ Real-time rendering speed (up to 76 FPS)

Comparison with SOTA

Want to drag the slider? Visit our project page for the full interactive comparison. 😄

LDR Comparison

Computer	Luckycat	Flower

HDR Comparison

Sofa	Bear	Bathroom

Preparation

Requirements

• Python 3.10
• CUDA 12.1 or later
• PyTorch 2.1.2 (with CUDA 12.1 support)
• torchvision 0.16.2
• pytorch-lightning 1.4.2

Option 1: Docker (Recommended)

We provide a Docker image environment for easy setup:

docker pull zeldam1/zhm_docker:zhm-py310-torch21
docker run --gpus all -it -v /workdir:/workdir  --shm-size 64g zeldam1/zhm_docker:zhm-py310-torch21 /bin/bash

cd Arbi-3DGSR

pip install pyiqa==0.1.10 pytorch-lightning==1.4.2 torchmetrics==0.6.0 taming-transformers-rom1504 scikit-learn kornia==0.6 open_clip_torch==2.0.2 transformers==4.38.2 clip accelerate==1.12.0 submodules/simple-knn submodules/diff-gaussian-rasterization

Option 2: Conda Environment

# Clone the repository
git clone [email protected]:huimin-zeng/Arbi-3DGSR.git
cd Arbi-3DGSR

# Create conda environment
conda create -y -n Arbi-3DGSR python=3.10
conda activate Arbi-3DGSR

# Install PyTorch with CUDA 12.1
pip install torch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 --index-url https://download.pytorch.org/whl/cu121


# Install other dependencies
pip install pytorch-lightning==1.4.2 torchmetrics==0.6.0 open-clip-torch==2.0.2
pip install pyiqa==0.1.10 taming-transformers-rom1504 scikit-learn kornia==0.6 transformers==4.38.2 clip accelerate==1.12.0

# Install submodules
pip install submodules/diff-gaussian-rasterization
pip install submodules/simple-knn

TODO

• Environment
• Dataset preparation
• Code releasing

Citation

If you find this work useful, please give us a star 🌟 and consider citing our paper:

@misc{zeng2026physicallyinspiredgaussiansplatting,
      title={Physically Inspired Gaussian Splatting for HDR Novel View Synthesis}, 
      author={Huimin Zeng and Yue Bai and Hailing Wang and Yun Fu},
      year={2026},
      eprint={2603.28020},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2603.28020}, 
}

Acknowledgments

This repository builds upon excellent prior work:

• 3D Gaussian Splatting by Kerbl et al.
• GaussHDR by Liu et al.

We thank their authors for sharing these excellent works!

License

This project is licensed under the MIT License.

Contact

For questions and issues, please open an issue on GitHub or contact [email protected].