From Restoration to Reconstruction: Rethinking 3D Gaussian Splatting for Underwater Scenes

2025-09-22

AI Summary

Underwater image degradation severely impairs 3D reconstruction performance, while existing simplified physical models exhibit poor generalization in complex scenes. To address this, we propose R-Splatting—a novel framework that jointly optimizes underwater image restoration and 3D Gaussian Splatting in an end-to-end manner for the first time. Our key contributions are: (1) a multi-source underwater restoration module to enhance input view quality; (2) a lightweight illumination generator coupled with contrastive loss to model illumination-invariant features; and (3) an uncertainty-aware transparency optimization mechanism to improve training stability and geometric robustness. Evaluated on Seathru-NeRF and our newly introduced BlueCoral3D dataset, R-Splatting achieves state-of-the-art performance across rendering metrics (PSNR, SSIM) and depth reconstruction error, unifying high-fidelity visual rendering with precise geometric reconstruction.

Technology Category

• Computer Vision: 3D Computer Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Planning, Routing, and Scheduling: Replanning and Plan Repair

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Underwater image degradation poses significant challenges for 3D reconstruction, where simplified physical models often fail in complex scenes. We propose extbf{R-Splatting}, a unified framework that bridges underwater image restoration (UIR) with 3D Gaussian Splatting (3DGS) to improve both rendering quality and geometric fidelity. Our method integrates multiple enhanced views produced by diverse UIR models into a single reconstruction pipeline. During inference, a lightweight illumination generator samples latent codes to support diverse yet coherent renderings, while a contrastive loss ensures disentangled and stable illumination representations. Furthermore, we propose extit{Uncertainty-Aware Opacity Optimization (UAOO)}, which models opacity as a stochastic function to regularize training. This suppresses abrupt gradient responses triggered by illumination variation and mitigates overfitting to noisy or view-specific artifacts. Experiments on Seathru-NeRF and our new BlueCoral3D dataset demonstrate that R-Splatting outperforms strong baselines in both rendering quality and geometric accuracy.

Problem

• Addressing underwater image degradation challenges in 3D reconstruction
• Integrating image restoration with 3D Gaussian Splatting for better rendering
• Improving geometric fidelity while handling illumination variations and artifacts

Innovation

• Unified framework integrating underwater restoration with 3D Gaussian Splatting
• Lightweight illumination generator with contrastive loss for coherent renderings
• Uncertainty-aware opacity optimization to regularize training and reduce overfitting