Summary
A paper by Diego Gomez, Antoine Guédon, and Nissim Maruani (arXiv:2504.06850, April 2025) addresses a key limitation of 3D Gaussian Splatting (3DGS): while 3DGS revolutionized fast novel view synthesis, its opacity-based formulation makes surface extraction inherently difficult. The authors derive a principled occupancy field for Gaussian splatting and show how it can be used to extract highly accurate watertight meshes of complex scenes. The key contribution is the introduction of learnable oriented normals at each Gaussian element, moving beyond isotropic blob-like primitives to anisotropic, orientation-aware representations. The resulting method, called Gaussian Wrapping, establishes a new state of the art on the DTU and Tanks and Temples benchmarks, and can recover thin structures such as bicycle spokes that conventional Gaussian-based approaches struggle to reconstruct. This work bridges the gap between fast radiance-field rendering and high-fidelity surface reconstruction, making Gaussian splatting practical for mesh extraction tasks in computer vision.
Paper Overview
- Field: cs.CV
- Authors: Diego Gomez, Antoine Guédon, Nissim Maruani
- Published: 2025-04-09
- arXiv: 2504.06850
Abstract
3D Gaussian Splatting (3DGS) has revolutionized fast novel view synthesis, but its opacity-based formulation makes surface extraction inherently difficult. This paper derives a principled occupancy field for Gaussian splatting and shows how it can be used to extract highly accurate watertight meshes of complex scenes.The key contribution is the introduction of learnable oriented normals at each Gaussian element. The proposed method, Gaussian Wrapping, establishes a new state of the art on the DTU and Tanks and Temples benchmarks, and is able to recover thin structures such as bicycle spokes.
Key Points
- Derives a principled occupancy field from Gaussian splatting, enabling mesh extraction from 3DGS representations.
- Introduces learnable oriented normals per Gaussian element, improving geometric fidelity over opacity-based splatting.
- Produces watertight meshes of complex scenes.
- Achieves new state-of-the-art results on DTU and Tanks and Temples.
- Successfully reconstructs thin structures (e.g., bicycle spokes) that are typically missed.
---
*Automatically collected on 2025-04-10.*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/177169721