NimbusGS: Unified 3D Scene Reconstruction under Hybrid Weather

2026-03-28

AI Summary

This work addresses the challenge of multi-view 3D reconstruction under complex mixed-weather conditions, where existing methods—typically designed for single adverse weather types—struggle to generalize. The authors propose a unified self-supervised framework that, for the first time, disentangles mixed-weather degradation into two components: a globally consistent transmission field modeling static atmospheric attenuation and view-specific particulate residuals capturing dynamic weather effects. To stabilize optimization under severe visibility degradation, they introduce a geometry-guided gradient scaling mechanism that enhances the robustness of 3D Gaussian representations. Evaluated across diverse mixed-weather scenarios, the method significantly outperforms state-of-the-art specialized approaches, achieving superior geometric accuracy and structural fidelity in reconstructed scenes.

Abstract

We present NimbusGS, a unified framework for reconstructing high-quality 3D scenes from degraded multi-view inputs captured under diverse and mixed adverse weather conditions. Unlike existing methods that target specific weather types, NimbusGS addresses the broader challenge of generalization by modeling the dual nature of weather: a continuous, view-consistent medium that attenuates light, and dynamic, view-dependent particles that cause scattering and occlusion. To capture this structure, we decompose degradations into a global transmission field and per-view particulate residuals. The transmission field represents static atmospheric effects shared across views, while the residuals model transient disturbances unique to each input. To enable stable geometry learning under severe visibility degradation, we introduce a geometry-guided gradient scaling mechanism that mitigates gradient imbalance during the self-supervised optimization of 3D Gaussian representations. This physically grounded formulation allows NimbusGS to disentangle complex degradations while preserving scene structure, yielding superior geometry reconstruction and outperforming task-specific methods across diverse and challenging weather conditions. Code is available at https://github.com/lyy-ovo/NimbusGS.

Problem

• 3D scene reconstruction
• adverse weather
• multi-view inputs
• weather degradation
• unified framework

Innovation

• 3D scene reconstruction
• adverse weather
• transmission field
• geometry-guided gradient scaling
• Gaussian splatting