Ref-GS : Directional Factorization for
2D Gaussian Splatting


1Huazhong University of Science and Technology 2University of Tübingen, Tübingen AI Center
3Westlake University 4Zhejiang University

Abstract

In this paper, we introduce Ref-GS, a novel approach for directional light factorization in 2D Gaussian splatting, which enables photorealistic view-dependent appearance rendering and precise geometry recovery. Ref-GS builds upon the deferred rendering of Gaussian splatting and applies directional encoding to the deferred-rendered surface, effectively reducing the ambiguity between orientation and viewing angle. Next, we introduce a spherical mip-grid to capture varying levels of surface roughness, enabling roughness-aware Gaussian shading. Additionally, we propose a simple yet efficient geometry-lighting factorization that connects geometry and lighting via the vector outer product, significantly reducing renderer overhead when integrating volumetric attributes. Our method achieves superior photorealistic rendering for a range of open-world scenes while also accurately recovering geometry.

Results and comparisons

Here we demostrate side-by-side videos comparing our method to top-performing baselines across different captured scenes.

Select a scene and a baseline method below:


Interactive visualization. Hover or tap to move the split.
Note how our method synthesizes accurate reflections of trees that move smoothly across the car surface and windows, while Gaussian-based methods produce blurry reflections that appear and disappear inconsistently with changes in viewpoint.
Note that our model renders reflections with more high-frequency details and is capable of producing realistic near-field content reflections. These include clear reflections of tree branches and buildings in the sphere.
In this relatively diffuse scene with no significant view-dependent appearance changes, we achieved a successful reconstruction of the geometries of the "reflective table center".
Although semi-transparent but reflective surfaces (such as windows) present certain challenges for our method, it is still capable of simulating convincing reflection effects on windshields.

Ablation study

Here we demostrate side-by-side videos comparing our full method to versions of our method where key components have been ablated. See more details in the paper.

Select an ablation below:


Interactive visualization. Hover or tap to move the split.
Using the G-buffer instead of Sph-Mip (i.e. w/o Sph-Mip Encoding) or without deferred shading (i.e., w/o Deferred Shading), sharp details, such as tree branches reflected in the sphere, are not accurately reconstructed. It is necessary to use multi-level spherical feature grid strategies (i.e., w/o Mipmap), otherwise rough surfaces will fail to be reconstructed and artifacts will appear during rendering. Additionally, directional factorization (i.e., w/o Directional Factorization) is essential for modeling near-field inter-reflections.
Using the G-buffer instead of Sph-Mip (i.e. w/o Sph-Mip Encoding) or without deferred shading (i.e., w/o Deferred Shading), sharp details, such as tree branches reflected in the sphere, are not accurately reconstructed. It is necessary to use multi-level spherical feature grid strategies (i.e., w/o Mipmap), otherwise rough surfaces will fail to be reconstructed and artifacts will appear during rendering. Additionally, directional factorization (i.e., w/o Directional Factorization) is essential for modeling near-field inter-reflections.

Shape Reconstruction

Here we demostrate side-by-side videos comparing our method to top-performing baselines across different captured scenes.

Select a reconstruction below:


Interactive visualization. Hover or tap to move the split.

Synthetic results

Here we demostrate side-by-side videos comparing our method to top-performing baselines across different captured scenes.

Select a scene and a baseline method below:


Interactive visualization. Hover or tap to move the split.

Related works

  • Ref-NeRF: Structured View-Dependent Appearance for Neural Radiance Fields
  • GaussianShader: 3D Gaussian Splatting with Shading Functions for Reflective Surfaces
  • 3iGS: Factorised Tensorial Illumination for 3D Gaussian Splatting
  • 3D Gaussian Splatting with Deferred Reflection
  • 2D Gaussian Splatting for Geometrically Accurate Radiance Fields