Reproduce “RT-GS: Gaussian Splatting with Reflection and Transmittance Primitives” 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
Kunnong Zeng, Chensheng Peng, Yichen Xie, Masayoshi Tomizuka, Cem Yuksel
RT-GS introduces a unified Gaussian Splatting framework that simultaneously reconstructs complex specular reflections and the appearance of objects through semi-transparent surfaces. By integrating a microfacet material model and differentiable ray tracing, the method produces sharper reflections and accurately models internal and external views through transparent objects, achieving competitive quantitative metrics on reflection and transparent scene datasets.
Key insights
- Separate Gaussian primitives for diffuse, reflection, and transmittance, coupled with a microfacet material model, enable a unified framework to handle diverse light interactions simultaneously.
- Differentiable ray tracing on specialized Gaussians, guided by an extracted surface mesh, is crucial for accurately capturing view-dependent specular reflections and two-bounce light paths through transparent objects.
- Using the full microfacet model without the split-sum approximation is essential for proper joint optimization and preventing degradation of diffuse surface training, leading to better separation of diffuse and specular components.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.