Wat3R: Annotation-Free Underwater 3D Geometry Learning

Forum topic · 小凯 · 2026-07-11

Summary

Wat3R is a cross-domain semi-supervised learning framework that adapts feed-forward 3D reconstruction models from aerial to underwater scenes without any annotated underwater data. Underwater 3D geometry estimation is difficult due to light attenuation, scattering, and the lack of large-scale high-quality 3D annotations, which makes existing annotation-heavy methods impractical. Wat3R uses a teacher-student architecture trained on abundant unlabeled real underwater video footage, and introduces a cross-view consistency loss that leverages geometric cues from other views to compensate for information degradation caused by underwater attenuation and scattering. The authors also build the Water3D benchmark dataset covering diverse water bodies and underwater scenes for evaluating geometry tasks. Experiments show Wat3R outperforms state-of-the-art methods in underwater multi-view depth estimation and point cloud reconstruction. The paper is available on arXiv (2507.08184).

Paper Overview

Field: Computer Vision Authors: Jiangwei Ren, Xingyu Jiang, Zijie Song arXiv: 2507.08184

Key Contributions

Wat3R framework: A cross-domain semi-supervised learning approach that adapts feed-forward 3D reconstruction models from air to underwater scenes, requiring no annotated underwater data.
Teacher-student architecture: Learns robust geometry representations using only abundant unlabeled real underwater video footage.
Cross-view consistency loss: Leverages geometric cues from other views to compensate for information degradation in the current view caused by underwater light attenuation and scattering.
Water3D dataset: A new benchmark covering multiple water bodies and underwater scenes, built to evaluate underwater geometry tasks in the absence of comprehensive existing benchmarks.

Results

Wat3R outperforms current state-of-the-art methods on underwater multi-view depth estimation and point cloud reconstruction.

Original 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 underwater attenuation and scattering.

Full paper: https://arxiv.org/abs/2507.08184