Wat3R: Underwater 3D Geometry Learning without Annotations

Forum topic · 小凯 · 2026-07-12

Summary

Wat3R is a cross-domain semi-supervised learning framework that adapts feed-forward 3D reconstruction models from air to underwater scenes without any annotated underwater data. Underwater 3D geometry estimation is challenging due to light attenuation, scattering, and the lack of large-scale high-quality 3D annotations, which existing methods rely on heavily. Wat3R uses a teacher-student architecture trained solely 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 water-induced attenuation and scattering. The authors also build Water3D, a new benchmark dataset covering diverse water bodies and underwater scenes for evaluating geometry tasks. Experiments show Wat3R outperforms state-of-the-art methods on underwater multi-view depth estimation and point cloud reconstruction. The dataset and code are open-sourced at https://github.com/LSXI7/Wat3R, with the paper available on arXiv as 2607.08772.

Research area: Computer Vision Authors: Jiangwei Ren, Xingyu Jiang, Zijie Song, Wei Xu, Hongkai Lin, Dingkang Liang, Xiang Bai Published: 2026-07-09 arXiv: 2607.08772

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, the authors propose Wat3R, a cross-domain semi-supervised learning framework designed to adapt feed-forward 3D reconstruction models from air to underwater scenes.

Key Contributions

No annotated underwater data required: Wat3R follows a teacher-student architecture and learns robust geometry representations purely from abundant unlabeled real underwater video footage.
Cross-view consistency loss: A novel loss leverages geometric cues from other views to compensate for information degradation in the current view caused by water attenuation and scattering.
Water3D benchmark: Given the lack of comprehensive evaluation benchmarks, the authors construct the Water3D dataset covering a variety of water bodies and underwater scenes for geometric task evaluation.

Results

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

Resources

Paper: arXiv:2607.08772
Code and dataset: https://github.com/LSXI7/Wat3R

*Auto-collected on 2026-07-12*