Swimm3R: Splatting with Medium-aware SfM for Underwater 3D Reconstruction

2026-08-01

AI Summary

This work addresses the failure of structure-from-motion in underwater 3D reconstruction caused by light scattering and attenuation. The authors propose Underwater Beta Splatting, a novel approach that integrates medium-aware structure-from-motion with an extension of Gaussian splatting based on Beta primitives. By leveraging a feedforward network to extract geometric priors from in-air images and jointly optimizing camera poses, medium parameters, and point cloud geometry through a physics-driven underwater imaging model, the method enables robust reconstruction. A key innovation is the introduction of a scattering-aware geometric gradient, which stabilizes underwater geometry representation. Evaluated on the newly introduced Barbados underwater video dataset, the approach significantly improves reconstruction quality—yielding an average PSNR gain of 1.47 dB—and enhances downstream localization performance, with RRA@15 and RTA@15 increasing by 2.0 and 2.4 percentage points, respectively.

Abstract

We propose Swimm3R, a unified framework that combines medium-aware structure-from-motion (SfM) with Underwater Beta Splatting to address scattering- and attenuation-induced failures in underwater 3D reconstruction. Swimm3R distills in-air geometric priors into a feed-forward backbone and uses a physics head to regress underwater image-formation parameters, camera poses, and restored point clouds. Additionally, we introduce Underwater Beta Splatting, which extends Gaussian splatting with Beta primitives and scattering-aware geometric gradients for stable underwater geometry representation. We further establish the Barbados underwater video dataset to demonstrate the effectiveness of our method in challenging underwater environments. On this dataset, Swimm3R robustly recovers underwater scene structure under challenging scattering conditions, yielding coherent seafloor geometry. Using these predicted point clouds, the proposed Underwater Beta Splatting improves average PSNR by $1.47$ dB over WaterSplatting while increasing downstream localization performance by $2.0$ and $2.4$ percentage points in RRA@15 and RTA@15, respectively.

Problem

• underwater 3D reconstruction
• scattering
• attenuation
• structure-from-motion
• Gaussian splatting

Innovation

• Underwater 3D Reconstruction
• Medium-aware SfM
• Beta Splatting
• Scattering-aware Geometry
• Gaussian Splatting