HiRes-GS: Hierarchical Resolution Scaling for Sparse-View High-Resolution 3D Gaussian Splatting

Qinzheng Zhou, Zaychik Liu, Lijing Lu, Zhihang Li

Keywords: 3D Gaussian Splatting, Multi-resolution

TL;DR: We propose HiRes-GS, a hierarchical multi-scale paradigm that reconstructs scenes through a coarse-to-fine hierarchical optimization process‌.

Abstract: Sparse-view 3D Gaussian Splatting (3DGS) reconstructs scenes using 3D Gaussians from sparse input views. Yet, this method is prone to overfitting, which is exacerbated at higher resolutions as the expanded dimensionality amplifies floating artifacts and reconstruction ambiguities. In this paper, we systematically investigate the reconstruction performance of 3DGS in sparse views under varying resolutions‌, and we are the first to discover that low-resolution inputs yield stable global structures but lose fine details, while high-resolution inputs capture richer local features at the cost of increased noise and ghosting. Motivated by this finding, we further propose HiRes-GS, a hierarchical multi-scale paradigm that reconstructs scenes through a coarse-to-fine hierarchical optimization process‌. Our approach employs a matching-based pruning strategy to anchor high-resolution reconstructions to stabilize structural priors and filtering noise through cross-scale consistency, and a multi-scale pseudo-view regularization to refine local details without amplifying noise. Extensive experiments on the LLFF and Mip-NeRF360 datasets demonstrate that HiRes-GS significantly outperforms existing methods, particularly under demanding high-resolution conditions. ‌Moreover, our paradigm can be seamlessly integrated into other 3DGS-based pipelines, thereby extending the field from low-resolution reconstructions to high-fidelity outputs under real-world sparse-view constraints.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 6791