Robust and Efficient 3D Gaussian Splatting for Urban Scene Reconstruction

TL;DR

This paper introduces a robust and efficient framework for urban scene reconstruction and real-time rendering using 3D Gaussian splatting, incorporating scene partitioning, visibility-based image selection, level-of-detail control, and appearance transformation modules.

Contribution

It presents a novel combination of techniques for fast, robust, and high-fidelity urban scene reconstruction with real-time rendering capabilities.

Findings

Outperforms previous methods in efficiency and quality.

Enables real-time rendering of large-scale urban scenes.

Effectively handles appearance variations across multi-view images.

Abstract

We present a framework that enables fast reconstruction and real-time rendering of urban-scale scenes while maintaining robustness against appearance variations across multi-view captures. Our approach begins with scene partitioning for parallel training, employing a visibility-based image selection strategy to optimize training efficiency. A controllable level-of-detail (LOD) strategy explicitly regulates Gaussian density under a user-defined budget, enabling efficient training and rendering while maintaining high visual fidelity. The appearance transformation module mitigates the negative effects of appearance inconsistencies across images while enabling flexible adjustments. Additionally, we utilize enhancement modules, such as depth regularization, scale regularization, and antialiasing, to improve reconstruction fidelity. Experimental results demonstrate that our method effectively reconstructs urban-scale scenes and outperforms previous approaches in both efficiency and quality. The source code is available at: https://yzslab.github.io/REUrbanGS.