Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

7DGS: Unified Spatial-Temporal-Angular Gaussian Splatting

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

TL;DR

7DGS introduces a unified spatio-temporal-angular Gaussian representation that encodes geometry, dynamics, and view-dependent appearance in 7D Gaussians. A principled conditional slicing mechanism projects these into time- and direction-conditioned 3D Gaussians, preserving real-time performance and compatibility with existing 3D Gaussian Splatting pipelines. An adaptive Gaussian refinement module further tunes parameters via neural residuals to handle complex deformations. Across D-NeRF, Technicolor, and a new 7DGS-PBR dataset, 7DGS achieves up to 7.36 dB PSNR improvements and real-time rendering speeds (over 401 FPS), outperforming state-of-the-art methods while using substantially fewer Gaussian primitives. This unified approach enables accurate modeling of moving highlights and time-varying anisotropy, advancing practical dynamic scene synthesis and rendering.

Abstract

Real-time rendering of dynamic scenes with view-dependent effects remains a fundamental challenge in computer graphics. While recent advances in Gaussian Splatting have shown promising results separately handling dynamic scenes (4DGS) and view-dependent effects (6DGS), no existing method unifies these capabilities while maintaining real-time performance. We present 7D Gaussian Splatting (7DGS), a unified framework representing scene elements as seven-dimensional Gaussians spanning position (3D), time (1D), and viewing direction (3D). Our key contribution is an efficient conditional slicing mechanism that transforms 7D Gaussians into view- and time-conditioned 3D Gaussians, maintaining compatibility with existing 3D Gaussian Splatting pipelines while enabling joint optimization. Experiments demonstrate that 7DGS outperforms prior methods by up to 7.36 dB in PSNR while achieving real-time rendering (401 FPS) on challenging dynamic scenes with complex view-dependent effects. The project page is: https://gaozhongpai.github.io/7dgs/.

7DGS: Unified Spatial-Temporal-Angular Gaussian Splatting

TL;DR

7DGS introduces a unified spatio-temporal-angular Gaussian representation that encodes geometry, dynamics, and view-dependent appearance in 7D Gaussians. A principled conditional slicing mechanism projects these into time- and direction-conditioned 3D Gaussians, preserving real-time performance and compatibility with existing 3D Gaussian Splatting pipelines. An adaptive Gaussian refinement module further tunes parameters via neural residuals to handle complex deformations. Across D-NeRF, Technicolor, and a new 7DGS-PBR dataset, 7DGS achieves up to 7.36 dB PSNR improvements and real-time rendering speeds (over 401 FPS), outperforming state-of-the-art methods while using substantially fewer Gaussian primitives. This unified approach enables accurate modeling of moving highlights and time-varying anisotropy, advancing practical dynamic scene synthesis and rendering.

Abstract

Real-time rendering of dynamic scenes with view-dependent effects remains a fundamental challenge in computer graphics. While recent advances in Gaussian Splatting have shown promising results separately handling dynamic scenes (4DGS) and view-dependent effects (6DGS), no existing method unifies these capabilities while maintaining real-time performance. We present 7D Gaussian Splatting (7DGS), a unified framework representing scene elements as seven-dimensional Gaussians spanning position (3D), time (1D), and viewing direction (3D). Our key contribution is an efficient conditional slicing mechanism that transforms 7D Gaussians into view- and time-conditioned 3D Gaussians, maintaining compatibility with existing 3D Gaussian Splatting pipelines while enabling joint optimization. Experiments demonstrate that 7DGS outperforms prior methods by up to 7.36 dB in PSNR while achieving real-time rendering (401 FPS) on challenging dynamic scenes with complex view-dependent effects. The project page is: https://gaozhongpai.github.io/7dgs/.

Paper Structure

This paper contains 13 sections, 17 equations, 3 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminary
  4. Our Approach
  5. 7D Gaussian Representation
  6. Conditional Slicing Mechanism
  7. Adaptive Gaussian Refinement
  8. Optimization and Rendering Pipeline
  9. Experiments
  10. Experimental Protocol
  11. Comparison with Baseline
  12. Comparison with State-of-the-Art
  13. Conclusion

Figures (3)

  • Figure 1: Visualization of volumetric rendering for dynamic scenes. Top-left: Our 7DGS rendering. Bottom-left: Physically-based rendering via ray/path tracing (note: floating artifacts in the heart scene are caused by incomplete segmentation in CT scans and are not rendering artifacts). Right: Comparison between our method and 4DGS in highlighted red regions.
  • Figure 2: Proposed 7DGS compatible with the existing 3DGS pipeline.
  • Figure 3: Qualitative comparison of methods on the 7DGS-PBR, D-NeRF pumarola2021d, and Technicolor sabater2017dataset datasets (zoom in for details).