Wat3R: Underwater 3D Geometry Learning without Annotations

2026-07-09

AI Summary

This work addresses the significant challenges of 3D reconstruction in underwater environments, where severe light attenuation, scattering, and the absence of high-quality 3D annotations hinder performance. To overcome these limitations, the authors propose Wat3R, a cross-domain semi-supervised learning framework that leverages a teacher–student architecture to transfer a 3D reconstruction model pretrained in air to underwater scenes. Wat3R learns geometric representations exclusively from unlabeled underwater video sequences and introduces a cross-view consistency loss to mitigate information degradation inherent in single-view observations. Notably, this study achieves the first geometry-aware learning for underwater 3D reconstruction without requiring any underwater 3D annotations and introduces Water3D, the first comprehensive benchmark dataset for underwater 3D tasks. Experiments demonstrate that Wat3R substantially outperforms existing methods in multi-view depth estimation and point cloud reconstruction.

Abstract

Estimating 3D geometry in underwater environments presents unique challenges due to light attenuation, scattering, and the absence of large-scale, high-quality 3D annotations. Pioneering methods rely on massive dense annotations that are impractical in underwater settings. In this paper, we propose Wat3R, a cross-domain semi-supervised learning framework designed to adapt feed-forward 3D reconstruction models from air to underwater scenes. Uniquely, our method eliminates the need for any annotated underwater data following a teacher-student architecture, that learns robust geometry representations merely on abundant unlabeled real underwater video footage. We also design a cross-view consistency loss that leverages geometric cues from other views to compensate for the information degradation in the current view caused by water attenuation and scattering. Furthermore, considering the lack of comprehensive evaluation benchmarks, we construct Water3D, a diverse dataset covering various water bodies and underwater scenarios, designed for geometric task evaluation. Experimental results demonstrate that Wat3R outperforms current state-of-the-art methods in underwater multi-view depth estimation and point cloud reconstruction. The dataset and code are available at https://github.com/LSXI7/Wat3R .

Problem

• underwater 3D geometry
• light attenuation
• scattering
• 3D annotations
• multi-view depth estimation

Innovation

• underwater 3D reconstruction
• semi-supervised learning
• teacher-student architecture
• cross-view consistency
• annotation-free