Sea-Undistort: A Dataset for Through-Water Image Restoration in High Resolution Airborne Bathymetric Mapping
Maximilian Kromer, Panagiotis Agrafiotis, Begüm Demir
TL;DR
Sea-Undistort provides $1200$ paired $512\times512$ through-water scenes with distortion-free and distorted views modeled with sun glint, wave distortions, and scattering to enable supervised restoration for aerial shallow-water bathymetry. The authors benchmark diffusion-based restoration models (NDR-Restore, ResShift) and an enhanced ResShift+EF that uses an early-fusion sun-glint mask, demonstrating improved seabed visibility and DSM completeness when integrated with SfM-MVS and learning-based SDB. ResShift+EF achieves the best perceptual restoration while maintaining bathymetric accuracy, highlighting the value of synthetic paired data for real-world bathymetric pipelines. The work provides public data, weights, and code, enabling broader adoption and further improvements in through-water image restoration and high-resolution seabed mapping.
Abstract
Accurate image-based bathymetric mapping in shallow waters remains challenging due to the complex optical distortions such as wave induced patterns, scattering and sunglint, introduced by the dynamic water surface, the water column properties, and solar illumination. In this work, we introduce Sea-Undistort, a comprehensive synthetic dataset of 1200 paired 512x512 through-water scenes rendered in Blender. Each pair comprises a distortion-free and a distorted view, featuring realistic water effects such as sun glint, waves, and scattering over diverse seabeds. Accompanied by per-image metadata such as camera parameters, sun position, and average depth, Sea-Undistort enables supervised training that is otherwise infeasible in real environments. We use Sea-Undistort to benchmark two state-of-the-art image restoration methods alongside an enhanced lightweight diffusion-based framework with an early-fusion sun-glint mask. When applied to real aerial data, the enhanced diffusion model delivers more complete Digital Surface Models (DSMs) of the seabed, especially in deeper areas, reduces bathymetric errors, suppresses glint and scattering, and crisply restores fine seabed details. Dataset, weights, and code are publicly available at https://www.magicbathy.eu/Sea-Undistort.html.