Research map / 3D vision / Gaussian splatting
Direct Conditional Parameterization for N-Dimensional Splatting
Zhongpai Gao, Benjamin Planche, Van N Nguyen, Meng Zheng, Anwesa Choudhuri, Arun Innanje, Terrence Chen, Ziyan Wu
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes Direct Conditional Parameterization for N-dimensional Gaussian splatting (N-DGS), which conditions lifted joint Gaussians over position and conditioning variables (view, time) to get view- and time-dependent primitives.
Problem
Modeling view-dependent and dynamic appearance with Gaussian splatting by conditioning primitives on variables such as view direction and time, where the abstract indicates the prior lifted-distribution formulation is the setting being improved.
Approach
Each primitive is lifted into a joint Gaussian over 3D position and conditioning variables and conditionally sliced at render time, so mean, opacity and covariance vary continuously with view or time; the title suggests the conditional parameterization is defined directly rather than via a separate slicing step (details not in the abstract).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace covariance-derived conditioning with direct conditional parameterization (Gaussian) | N-DGS · Mip-NeRF 360 (MCMC) | PSNR (dB): 26.88 -> 27.67 | Improved |
| replace covariance-derived conditioning with direct conditional parameterization (Gaussian) | N-DGS · D-NeRF (MCMC) | PSNR (dB): 34.63 -> 34.64 | No clear effect |
| replace covariance-derived conditioning with direct conditional parameterization (Beta) | UBS · 7DGS-PBR (MCMC) | PSNR (dB): 30.61 -> 31.87 | Improved |
| replace covariance-derived conditioning with direct conditional parameterization (Beta) | UBS · NeRF Synthetic (MCMC) | PSNR (dB): 34.48 -> 34.53 | No clear effect |
| direct conditioning with fused CUDA slicing | N-DGS/UBS · Slicing time, 1M primitives | ms: 0.71 -> 0.10 (C=3) | Improved |
| add learned per-dimension coupling Lambda instead of fixed Lambda=0 (ablation) | dGS · 6DGS-PBR | PSNR (dB): 36.54 -> 37.75 | Improved |
| add learned per-dimension coupling Lambda instead of fixed Lambda=1 (ablation) | dGS · 6DGS-PBR | PSNR (dB): 37.31 -> 37.75 | Improved |
| add learned Lambda instead of fixed Lambda=0 (ablation) | dGS · NeRF Synthetic | PSNR (dB): 33.84 -> 33.91 | Improved |
| add spatial scaling s-bar on the coupling factor (ablation) | dGS · NeRF Synthetic | PSNR (dB): 31.26 -> 33.91 | Improved |
| add spatial scaling s-bar on the coupling factor (ablation) | dGS · 6DGS-PBR | PSNR (dB): 34.78 -> 37.75 | Improved |
| use structured factorized regression instead of unrestricted direct regression (ablation) | dGS · NeRF Synthetic | PSNR (dB): 32.91 -> 33.91 | Improved |
| use full coupling (Lambda=1) instead of opacity-only (Lambda=0) (ablation) | dGS · NeRF Synthetic vs 6DGS-PBR | PSNR (dB): NeRF Synthetic 33.84 -> 33.78; 6DGS-PBR 36.54 -> 37.31 | Mixed |
Experimental setup
Builds on: 3D Gaussian Splatting N-dimensional Gaussians
Closest papers by meaning
- 3D Student Splatting and Scooping (CVPR 2025 · similarity 0.89)
- ReorgGS: Equivalent Distribution Reorganization for 3D Gaussian Splatting (NeurIPS 2026 · similarity 0.88)
- Fully Explicit Dynamic Gaussian Splatting (NeurIPS 2024 · similarity 0.88)
- HybridGS: Decoupling Transients and Statics with 2D and 3D Gaussian Splatting (CVPR 2025 · similarity 0.88)
- Spec-Gaussian: Anisotropic View-Dependent Appearance for 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.88)
- 6DGS: Enhanced Direction-Aware Gaussian Splatting for Volumetric Rendering (ICLR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
