WaterFlow: Explicit Physics-Prior Rectified Flow for Underwater Saliency Mask Generation

Runting LI; Shijie LIAN; Hua LI; Yutong LI; Wenhui WU; Sam KWONG

doi:10.1109/ICASSP55912.2026.11461353

WaterFlow: Explicit Physics-Prior Rectified Flow for Underwater Saliency Mask Generation

Runting LI
, Shijie LIAN
, Hua LI^*
, Yutong LI
, Wenhui WU
, Sam KWONG

^*Corresponding author for this work

Lingnan University

Research output: Book Chapters | Papers in Conference Proceedings › Conference paper (refereed) › Referred Conference Paper › peer-review

Abstract

Underwater Salient Object Detection (USOD) faces significant challenges, including underwater image quality degradation and domain gaps. Existing methods tend to ignore the physical principles of underwater imaging or simply treat degradation phenomena in underwater images as interference factors that must be eliminated, failing to fully exploit the valuable information they contain. We propose WaterFlow, a rectified flow-based framework for underwater salient object detection that innovatively incorporates underwater physical imaging information as explicit priors directly into the network training process and introduces temporal dimension modeling, significantly enhancing the model’s capability for salient object identification. On the USOD10K dataset, WaterFlow achieves a 0.072 gain in Sm, demonstrating the effectiveness and superiority of our method. https://github.com/Theo-polis/WaterFlow.

Original language	English
Title of host publication	ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Proceedings
Publisher	IEEE
Pages	10972-10976
Number of pages	5
ISBN (Electronic)	9798331567019
ISBN (Print)	9798331567026
DOIs	https://doi.org/10.1109/ICASSP55912.2026.11461353
Publication status	Published - 2026
Event	2026 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2026 - Barcelona, Spain Duration: 4 May 2026 → 8 May 2026

Publication series

Name	Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing
Publisher	IEEE
ISSN (Print)	1520-6149
ISSN (Electronic)	2379-190X

Conference

Conference	2026 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2026
Abbreviated title	ICASSP 2026
Country/Territory	Spain
City	Barcelona
Period	4/05/26 → 8/05/26

Funding

This work was supported in part by the National Natural Science Foundation of China under Grant 62461018, 62376162; in part by the Hainan Provincial Natural Science Foundation of China under Grant No. 625YXQN594; in part by the Innovation Platform for ”New Star of South China Sea” of Hainan Province under Grant No. NHXXRCXM202306; in part by the Research Grants Council of the Hong Kong Special Administrative Region, China under Grant STG5/E-103/24-R.

Keywords

Underwater Salient Object Detection
Rectified Flow
Physical prior
Generative model

Access to Document

10.1109/ICASSP55912.2026.11461353

Tomorrow is Now: Eco-friendly Autonomous Construction through Sustainable 3D Concrete Printing (LU Part)
LIEW, K. M. (PI), DAI, J. G. (CoPI), LIAO, W. H. (CoPI), NG, S. T. T. (CoPI), ZHANG, X. (CoPI), YE, H. (CoPI), WENG, Y. (CoI), QIU, J. (CoPI), YIN, B. (CoPI), KWONG, S. T. W. (CoPI) & RABCZUK, T. (CoPI)
Research Grants Council (Hong Kong, China)
1/01/25 → 31/12/29
Project: Grant Research

Cite this

LI, R., LIAN, S., LI, H., LI, Y., WU, W., & KWONG, S. (2026). WaterFlow: Explicit Physics-Prior Rectified Flow for Underwater Saliency Mask Generation. In ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Proceedings (pp. 10972-10976). (Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing). IEEE. https://doi.org/10.1109/ICASSP55912.2026.11461353

LI, Runting ; LIAN, Shijie ; LI, Hua et al. / WaterFlow: Explicit Physics-Prior Rectified Flow for Underwater Saliency Mask Generation. ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Proceedings. IEEE, 2026. pp. 10972-10976 (Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing).

@inproceedings{a221561a3f0f4e029db32886bfc496cb,

title = "WaterFlow: Explicit Physics-Prior Rectified Flow for Underwater Saliency Mask Generation",

abstract = "Underwater Salient Object Detection (USOD) faces significant challenges, including underwater image quality degradation and domain gaps. Existing methods tend to ignore the physical principles of underwater imaging or simply treat degradation phenomena in underwater images as interference factors that must be eliminated, failing to fully exploit the valuable information they contain. We propose WaterFlow, a rectified flow-based framework for underwater salient object detection that innovatively incorporates underwater physical imaging information as explicit priors directly into the network training process and introduces temporal dimension modeling, significantly enhancing the model{\textquoteright}s capability for salient object identification. On the USOD10K dataset, WaterFlow achieves a 0.072 gain in Sm, demonstrating the effectiveness and superiority of our method. https://github.com/Theo-polis/WaterFlow.",

keywords = "Underwater Salient Object Detection, Rectified Flow, Physical prior, Generative model",

author = "Runting LI and Shijie LIAN and Hua LI and Yutong LI and Wenhui WU and Sam KWONG",

year = "2026",

doi = "10.1109/ICASSP55912.2026.11461353",

language = "English",

isbn = "9798331567026",

series = "Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing",

publisher = "IEEE",

pages = "10972--10976",

booktitle = "ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Proceedings",

note = "2026 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2026, ICASSP 2026 ; Conference date: 04-05-2026 Through 08-05-2026",

}

LI, R, LIAN, S, LI, H, LI, Y, WU, W & KWONG, S 2026, WaterFlow: Explicit Physics-Prior Rectified Flow for Underwater Saliency Mask Generation. in ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Proceedings. Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing, IEEE, pp. 10972-10976, 2026 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2026, Barcelona, Spain, 4/05/26. https://doi.org/10.1109/ICASSP55912.2026.11461353

WaterFlow: Explicit Physics-Prior Rectified Flow for Underwater Saliency Mask Generation. / LI, Runting; LIAN, Shijie; LI, Hua et al.
ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Proceedings. IEEE, 2026. p. 10972-10976 (Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing).

Research output: Book Chapters | Papers in Conference Proceedings › Conference paper (refereed) › Referred Conference Paper › peer-review

TY - GEN

T1 - WaterFlow: Explicit Physics-Prior Rectified Flow for Underwater Saliency Mask Generation

AU - LI, Runting

AU - LIAN, Shijie

AU - LI, Hua

AU - LI, Yutong

AU - WU, Wenhui

AU - KWONG, Sam

PY - 2026

Y1 - 2026

N2 - Underwater Salient Object Detection (USOD) faces significant challenges, including underwater image quality degradation and domain gaps. Existing methods tend to ignore the physical principles of underwater imaging or simply treat degradation phenomena in underwater images as interference factors that must be eliminated, failing to fully exploit the valuable information they contain. We propose WaterFlow, a rectified flow-based framework for underwater salient object detection that innovatively incorporates underwater physical imaging information as explicit priors directly into the network training process and introduces temporal dimension modeling, significantly enhancing the model’s capability for salient object identification. On the USOD10K dataset, WaterFlow achieves a 0.072 gain in Sm, demonstrating the effectiveness and superiority of our method. https://github.com/Theo-polis/WaterFlow.

AB - Underwater Salient Object Detection (USOD) faces significant challenges, including underwater image quality degradation and domain gaps. Existing methods tend to ignore the physical principles of underwater imaging or simply treat degradation phenomena in underwater images as interference factors that must be eliminated, failing to fully exploit the valuable information they contain. We propose WaterFlow, a rectified flow-based framework for underwater salient object detection that innovatively incorporates underwater physical imaging information as explicit priors directly into the network training process and introduces temporal dimension modeling, significantly enhancing the model’s capability for salient object identification. On the USOD10K dataset, WaterFlow achieves a 0.072 gain in Sm, demonstrating the effectiveness and superiority of our method. https://github.com/Theo-polis/WaterFlow.

KW - Underwater Salient Object Detection

KW - Rectified Flow

KW - Physical prior

KW - Generative model

U2 - 10.1109/ICASSP55912.2026.11461353

DO - 10.1109/ICASSP55912.2026.11461353

M3 - Conference paper (refereed)

SN - 9798331567026

T3 - Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing

SP - 10972

EP - 10976

BT - ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Proceedings

PB - IEEE

T2 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2026

Y2 - 4 May 2026 through 8 May 2026

ER -

LI R, LIAN S, LI H, LI Y, WU W, KWONG S. WaterFlow: Explicit Physics-Prior Rectified Flow for Underwater Saliency Mask Generation. In ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Proceedings. IEEE. 2026. p. 10972-10976. (Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing). doi: 10.1109/ICASSP55912.2026.11461353

WaterFlow: Explicit Physics-Prior Rectified Flow for Underwater Saliency Mask Generation

Abstract

Publication series

Conference

Funding

Keywords

Access to Document

Fingerprint

Projects

Tomorrow is Now: Eco-friendly Autonomous Construction through Sustainable 3D Concrete Printing (LU Part)

Cite this