Spatiotemporal Degradation-Aware 3D Gaussian Splatting for Realistic Underwater Scene Reconstruction

arXiv cs.CV / 4/28/2026

📰 NewsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper addresses the challenge of reconstructing realistic underwater scenes from video, arguing that current 3D reconstruction methods struggle due to both spatiotemporal imaging degradations (e.g., caustics, flickering, attenuation, backscattering).
It proposes MarineSTD-GS, a 3D Gaussian Splatting framework that explicitly models temporal and spatial degradation simultaneously for more realistic reconstructions.
The method uses paired Gaussian primitives—Intrinsic Gaussians for the true scene and Degraded Gaussians for the observed effects—where Degraded Gaussian colors are physically derived from Intrinsic ones via a Spatiotemporal Degradation Modeling (SDM) module.
To improve training stability and geometric accuracy, the authors introduce a Depth-Guided Geometry Loss and a Multi-Stage Optimization strategy, and evaluate on both simulated and real-world datasets.
The work also contributes a simulated benchmark covering diverse spatial/temporal degradations with ground-truth appearances, showing improved novel-view synthesis that better matches water-free scene appearance.

Abstract

Reconstructing realistic underwater scenes from underwater video remains a meaningful yet challenging task in the multimedia domain. The inherent spatiotemporal degradations in underwater imaging, including caustics, flickering, attenuation, and backscattering, frequently result in inaccurate geometry and appearance in existing 3D reconstruction methods. While a few recent works have explored underwater degradation-aware reconstruction, they often address either spatial or temporal degradation alone, falling short in more real-world underwater scenarios where both types of degradation occur. We propose MarineSTD-GS, a novel 3D Gaussian Splatting-based framework that explicitly models both temporal and spatial degradations for realistic underwater scene reconstruction. Specifically, we introduce two paired Gaussian primitives: Intrinsic Gaussians represent the true scene, while Degraded Gaussians render the degraded observations. The color of each Degraded Gaussian is physically derived from its paired Intrinsic Gaussian via a Spatiotemporal Degradation Modeling (SDM) module, enabling self-supervised disentanglement of realistic appearance from degraded images. To ensure stable training and accurate geometry, we further propose a Depth-Guided Geometry Loss and a Multi-Stage Optimization strategy. We also construct a simulated benchmark with diverse spatial and temporal degradations and ground-truth appearances for comprehensive evaluation. Experiments on both simulated and real-world datasets show that MarineSTD-GS robustly handles spatiotemporal degradations and outperforms existing methods in novel view synthesis with realistic, water-free scene appearances.