Title: Geometry Gaussians: Decoupling Appearance and Geometry in Gaussian Splatting

Original: arXiv:2606.05124v1 Announce Type: cross Abstract: After the success of 3D Gaussian Splatting (3DGS) for novel view synthesis, many works have explored how to also use it for geometric surface representation. However, extracting accurate geometric information directly from 3DGS remains challenging and can often reduce the appearance rendering quality. In this work, we show that 3DGS in its default form is inheritedly unsuited to represent texture and geometry at the same time, by training with complete ground-truth texture and geometry information. We also propose a simple solution by applying a single additional geometry opacity parameter to each splat, together with an optional transparency-curated optimization pipeline. Our experiments, both with ground-truth and vision foundation model geometric input, show that this change leads to improved rendering and geometry performance on a wide variety of dataset, and especially complex scenes with transparent objects benefit significantly from our method.

Rewritten:

Abstract: Following the widespread adoption of 3D Gaussian Splatting (3DGS) for novel view synthesis, numerous studies have investigated its potential for geometric surface reconstruction. Nevertheless, obtaining precise geometric data directly from 3DGS is difficult and frequently compromises the quality of rendered appearances. Through training with comprehensive ground-truth texture and geometry data, we demonstrate that standard 3DGS is inherently ill-equipped to simultaneously model both texture and geometry. To address this, we introduce a straightforward remedy: assigning an extra geometry-specific opacity parameter to each splat, supplemented by an optional optimization pipeline designed to handle transparency. Our evaluations, utilizing both ground-truth data and geometric inputs from vision foundation models, indicate that this modification enhances both rendering fidelity and geometric accuracy across diverse datasets. Notably, complex scenes featuring transparent objects experience substantial improvements thanks to our approach.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC