Shigon/BRDFusion_data

BRDFusion Preprocessed Data

This repository contains the preprocessed evaluation and training data for the paper BRDFusion: Physics Meets Generation for Urban Scene Inverse Rendering.

Project Page | GitHub | arXiv

BRDFusion combines physics-based inverse rendering with generative modeling for high-quality urban scene relighting. It decomposes urban videos into geometry, materials, and HDR lighting for novel-view rendering, relighting, and scene-edit applications.

Citation

@misc{liu2026brdfusion,
  title={BRDFusion: Physics Meets Generation for Urban Scene Inverse Rendering},
  author={Liu, Yi-Ruei and Lee, Jie-Ying and Huang, Zheng-Hui and Liu, Yu-Lun and Lin, Chih-Hao},
  year={2026}
}