Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

6DGS: Enhanced Direction-Aware Gaussian Splatting for Volumetric Rendering

Zhongpai Gao, Benjamin Planche, Meng Zheng, Anwesa Choudhuri, Terrence Chen, Ziyan Wu

TL;DR

6DGS advances real-time volumetric rendering by extending 3D Gaussian splatting with a 6D spatial-angular representation that captures view-dependent effects. It introduces conditional Gaussian slices, enhanced color via spherical harmonics, and direction-aware opacity control, while maintaining full compatibility with the 3DGS pipeline. The approach yields significant PSNR gains and substantial reductions in Gaussian points compared to 3DGS and N-DG, and it achieves high frame rates, including CUDA-accelerated variants. These results demonstrate robust performance across scenes with strong view-dependent lighting and generalization to scenes with weak view dependence, offering practical impact for AR/VR, gaming, and film production.

Abstract

Novel view synthesis has advanced significantly with the development of neural radiance fields (NeRF) and 3D Gaussian splatting (3DGS). However, achieving high quality without compromising real-time rendering remains challenging, particularly for physically-based ray tracing with view-dependent effects. Recently, N-dimensional Gaussians (N-DG) introduced a 6D spatial-angular representation to better incorporate view-dependent effects, but the Gaussian representation and control scheme are sub-optimal. In this paper, we revisit 6D Gaussians and introduce 6D Gaussian Splatting (6DGS), which enhances color and opacity representations and leverages the additional directional information in the 6D space for optimized Gaussian control. Our approach is fully compatible with the 3DGS framework and significantly improves real-time radiance field rendering by better modeling view-dependent effects and fine details. Experiments demonstrate that 6DGS significantly outperforms 3DGS and N-DG, achieving up to a 15.73 dB improvement in PSNR with a reduction of 66.5% Gaussian points compared to 3DGS. The project page is: https://gaozhongpai.github.io/6dgs/

6DGS: Enhanced Direction-Aware Gaussian Splatting for Volumetric Rendering

TL;DR

6DGS advances real-time volumetric rendering by extending 3D Gaussian splatting with a 6D spatial-angular representation that captures view-dependent effects. It introduces conditional Gaussian slices, enhanced color via spherical harmonics, and direction-aware opacity control, while maintaining full compatibility with the 3DGS pipeline. The approach yields significant PSNR gains and substantial reductions in Gaussian points compared to 3DGS and N-DG, and it achieves high frame rates, including CUDA-accelerated variants. These results demonstrate robust performance across scenes with strong view-dependent lighting and generalization to scenes with weak view dependence, offering practical impact for AR/VR, gaming, and film production.

Abstract

Novel view synthesis has advanced significantly with the development of neural radiance fields (NeRF) and 3D Gaussian splatting (3DGS). However, achieving high quality without compromising real-time rendering remains challenging, particularly for physically-based ray tracing with view-dependent effects. Recently, N-dimensional Gaussians (N-DG) introduced a 6D spatial-angular representation to better incorporate view-dependent effects, but the Gaussian representation and control scheme are sub-optimal. In this paper, we revisit 6D Gaussians and introduce 6D Gaussian Splatting (6DGS), which enhances color and opacity representations and leverages the additional directional information in the 6D space for optimized Gaussian control. Our approach is fully compatible with the 3DGS framework and significantly improves real-time radiance field rendering by better modeling view-dependent effects and fine details. Experiments demonstrate that 6DGS significantly outperforms 3DGS and N-DG, achieving up to a 15.73 dB improvement in PSNR with a reduction of 66.5% Gaussian points compared to 3DGS. The project page is: https://gaozhongpai.github.io/6dgs/
Paper Structure (28 sections, 8 equations, 8 figures, 9 tables, 1 algorithm)

This paper contains 28 sections, 8 equations, 8 figures, 9 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminary
  4. 6D Gaussian Representation
  5. Slice 6D Gaussian to Conditional 3DGS
  6. Method
  7. Theoretical Analysis of Conditional Gaussian
  8. Conditional Mean ($\mu_{\text{cond}}$).
  9. Conditional Covariance ($\Sigma_{\text{cond}}$).
  10. Conditional Opacity ($\alpha_{\text{cond}}$).
  11. Enhanced 6D Gaussian Representation
  12. Improved Control of Gaussians
  13. Compatibility with 3DGS
  14. Experiments
  15. Experimental Protocol
  16. ...and 13 more sections

Figures (8)

  • Figure 1: Visualizations of volumetric rendering. Top-left: our 6DGS rendering; bottom-right: physically-based rendering using ray/path tracing; right: comparison with 3DGS over the red regions.
  • Figure 2: Proposed method of direction-aware 6DGS compatible with the existing 3DGS pipeline. The position and opacity of the conditional 3D Gaussian are adjusted according to the view direction.
  • Figure 3: Qualitative comparison of methods on the 6DGS-PBR dataset (zoom in for details).
  • Figure 4: Qualitative comparison of methods on the subsurface scatter (SSS) dragon scene.
  • Figure 5: Effect of adjusting $\lambda_{\text{opa}}$ on the conditional probability density function (PDF) of the directional component $f_{\text{cond}}$, plotted as the functions of the Mahalanobis distance $D$ between the view direction $d$ and the Gaussian mean direction $\mu_d$.
  • ...and 3 more figures