Teaser image showing input images, novel view synthesis results, and surface reconstruction results.
TL;DR: We propose a new Gaussian Splatting representation under Atlanta world assumption for indoor and urban surface reconstruction.


How it works

In this paper, we propose an Atlanta-world guided implicit-structured Gaussian Splatting that achieves smooth indoor and urban scene reconstruction while preserving high-frequency details and rendering efficiency. By leveraging the Atlanta-world model, we ensure the accurate surface reconstruction for low-texture regions, while proposed novel implicit-structured GS representations provide smoothness without sacrificing efficiency and high-frequency details. Specifically, we propose a semantic GS representation to predict probability of all semantic regions and deploy a structure plane regularization with learnable plane indicators to global accurate surface reconstruction.







Comparisons to other methods





Method overview

A diagram explaining the method in broad strokes, like explained in the caption.
Given posed images and SfM points, we construct a sparse feature grid and represent scenes as implicit-structured Gaussians. Attributes are decoded using attribute decoder and semantic decoder, followed by rasterization and supervision with RGB images, monocular geometry priors, and semantic maps. To address multiview inconsistency in textureless regions, we introduce learnable explicit plane indicators based on the Atlanta world assumption. The indicators refine the global scene structure by regularizing Gaussian positions and orientations using 3D global planar and 2D local surface losses, ensuring alignment with structural elements such as walls, floors, and ceilings.