Reproduce “Flash-Splat: 3D Reflection Removal with Flash Cues and Gaussian Splats” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mingyang Xie, Haoming Cai, Sachin Shah, Yiran Xu, Brandon Y. Feng, Jia-Bin Huang, Christopher A. Metzler
Researchers at the University of Maryland and MIT present Flash-Splat, a system that achieves 3D transmission-reflection separation from transparent surfaces using unpaired multi-view flash and no-flash image sequences. The method leverages 3D Gaussian Splatting and integrates physical illumination cues to reconstruct distinct 3D representations, yielding significantly clearer reflection-free images and more accurate 3D scene understanding compared to prior techniques.
Key insights
- Integrating subtle physical cues from flash/no-flash illumination into a 3D inverse rendering pipeline, specifically with 3D Gaussian Splatting, enables robust reflection separation.
- The stringent requirement for perfectly paired flash and no-flash images can be overcome by leveraging multi-view sequences and synthesizing 'pseudo-paired' 2D and 3D representations.
- 3D Gaussian Splatting provides a suitable and efficient scene representation for this task, enabling high-quality 3D separation while mitigating artifacts often seen with other implicit neural representations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.