UW-GS: Distractor-Aware 3D Gaussian Splatting for Enhanced Underwater Scene Reconstruction

Haoran Wang, Nantheera Anantrasiricha, Fan Zhang, David Bull
School of Computer Science, University of Bristol, Bristol, UK
WACV 2025
Paper Code arXiv 🤗 Dataset

MY ALT TEXT


The diagram of our proposed UW-GS approach, combining a novel color appearance model, physically-based density control, and a binary motion mask to 3DGS. Our color appearance model uses view-direction R and depth z encoded by position encoding γ to estimate water condition parameters: attenuation factor TDi, backscatter factors TBi, and medium coefficients βdi, βbi, and bi. In the splatting process, the physical-based density control module addresses densification failures, and the binary motion mask handles distractors.

Abstract

3D Gaussian splatting (3DGS) offers the capability to achieve real-time high quality 3D scene rendering. However, 3DGS assumes that the scene is in a clear medium environment and struggles to generate satisfactory representations in underwater scenes, where light absorption and scattering are prevalent and moving objects are involved. To overcome these, we introduce a novel Gaussian Splatting-based method, UW-GS, designed specifically for underwater applications. It introduces a color appearance that models distance-dependent color variation, employs a new physics-based density control strategy to enhance clarity for distant objects, and uses a binary motion mask to handle dynamic content. Optimized with a well-designed loss function supporting scattering media and strengthened by pseudo-depth maps, UW-GS outperforms existing methods with PSNR gains up to 1.26dB. To fully verify the effectiveness of the model, we also developed a new underwater dataset, S-UW, with dynamic object masks.

Results

BibTeX

@inproceedings{wang2025uw,
  title={UW-GS: Distractor-aware 3d gaussian splatting for enhanced underwater scene reconstruction},
  author={Wang, Haoran and Anantrasirichai, Nantheera and Zhang, Fan and Bull, David},
  booktitle={2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  pages={3280--3289},
  year={2025},
  organization={IEEE}
}