README.md

---
tags:
- model_hub_mixin
- pytorch_model_hub_mixin
license: apache-2.0
base_model:
- facebook/VGGT-1B
pipeline_tag: image-to-3d
---

<h1 align="center">
  Wat3R: Underwater 3D Geometry Learning <br>
  without Annotations
</h1>

<div align="center">


[Jiangwei Ren](https://github.com/LSXI7),
[Xingyu Jiang](https://scholar.google.com/citations?user=h2W90MQAAAAJ&hl=en&oi=ao)<sup>†</sup>,
[Zijie Song](https://github.com/Sadak-X),
Wei Xu,
[Hongkai Lin](https://github.com/HongkLin),
[Dingkang Liang](https://dk-liang.github.io/)
and [Xiang Bai](https://scholar.google.com/citations?user=UeltiQ4AAAAJ&hl=en)

Huazhong University of Science & Technology.       
(†) Corresponding author.

</div>


<div align="center">

<a href="https://arxiv.org/abs/2607.08772"><img src="https://img.shields.io/badge/arXiv-2607.08772-b31b1b" alt='arxiv'></a>
<a href="https://huggingface.co/spaces/lsxi77777/Wat3R"><img src="https://img.shields.io/badge/%F0%9F%A4%97_Hugging_Face-Space-F0CD4B?labelColor=666EEE" alt='HuggingFace Space'></a>
<a href="https://www.apache.org/licenses/LICENSE-2.0"><img src="https://img.shields.io/badge/License-Apache--2.0-929292" alt='license'></a>
<a href="https://openxlab.org.cn/datasets/lsxi7/Water3D"><img src="https://img.shields.io/badge/OpenXLab-Dataset-blue" alt='data'></a>
<a href="https://huggingface.co/datasets/lsxi77777/Water3D"><img src="https://img.shields.io/badge/%F0%9F%A4%97_Hugging_Face-Dataset-F0CD4B?labelColor=666EEE" alt='HuggingFace Space'></a>

</div>

## 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.

## Citation

If you find our work useful in your research, please consider giving a star ⭐ and a citation

```bibtex
@inproceedings{ren2026wat3r,
  title={Wat3R: Underwater 3D Geometry Learning without Annotations},
  author={Ren, Jiangwei and Jiang, Xingyu and Song, Zijie and Xu, Wei and Lin, Hongkai and Liang, Dingkang and Bai, Xiang},
  booktitle={Proceedings of the European Conference on Computer Vision},
  year={2026}
}
```