GStex: Per-Primitive Texturing of 2D Gaussian Splatting for Decoupled Appearance and Geometry Modeling

📅 2024-09-19
🏛️ arXiv.org
📈 Citations: 1
Influential: 1
📄 PDF
🤖 AI Summary
Existing 2D Gaussian splatting tightly couples appearance and geometry within each Gaussian primitive, leading to inefficient texture modeling—even planar scenes require excessive primitives. This work proposes a decoupled modeling framework: each 2D Gaussian is assigned an independent, differentiable texture map, enabling the first primitive-level texture parameterization. By introducing UV coordinate mapping and jointly optimizing geometric and texture parameters, appearance representation is fully disentangled from geometric topology constraints. The method supports efficient retexturing, real-time editing, and cross-scene texture transfer. Experiments demonstrate that our approach achieves superior view synthesis quality and significantly improved planar texture fidelity—particularly in regions with complex textures—while using substantially fewer Gaussians than baseline methods.

Technology Category

Application Category

📝 Abstract
Gaussian splatting has demonstrated excellent performance for view synthesis and scene reconstruction. The representation achieves photorealistic quality by optimizing the position, scale, color, and opacity of thousands to millions of 2D or 3D Gaussian primitives within a scene. However, since each Gaussian primitive encodes both appearance and geometry, these attributes are strongly coupled--thus, high-fidelity appearance modeling requires a large number of Gaussian primitives, even when the scene geometry is simple (e.g., for a textured planar surface). We propose to texture each 2D Gaussian primitive so that even a single Gaussian can be used to capture appearance details. By employing per-primitive texturing, our appearance representation is agnostic to the topology and complexity of the scene's geometry. We show that our approach, GStex, yields improved visual quality over prior work in texturing Gaussian splats. Furthermore, we demonstrate that our decoupling enables improved novel view synthesis performance compared to 2D Gaussian splatting when reducing the number of Gaussian primitives, and that GStex can be used for scene appearance editing and re-texturing.
Problem

Research questions and friction points this paper is trying to address.

Decouples appearance and geometry modeling
Enhances visual quality with per-primitive texturing
Improves novel view synthesis with fewer primitives
Innovation

Methods, ideas, or system contributions that make the work stand out.

Per-primitive texturing for Gaussians
Decouples appearance and geometry modeling
Enhances novel view synthesis performance
🔎 Similar Papers
No similar papers found.
V
Victor Rong
University of Toronto, Vector Institute
J
Jingxiang Chen
University of Toronto, Vector Institute
Sherwin Bahmani
Sherwin Bahmani
University of Toronto
Computer VisionComputer GraphicsMachine Learning
K
Kiriakos N. Kutulakos
University of Toronto, Vector Institute
D
David B. Lindell
University of Toronto, Vector Institute