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

2025-10-14

AI Summary

Underwater salient object detection (USOD) suffers from severe image degradation and domain shift; existing methods typically treat degradation as noise, overlooking its underlying physical cues. To address this, we propose a physics-guided temporal diffusion model that explicitly incorporates underwater imaging priors—such as wavelength-dependent attenuation and backscattering—into the diffusion process via a rectified flow framework, ensuring physically consistent feature evolution. Additionally, we introduce temporal modeling across diffusion steps to enhance long-range saliency representation. Our approach overcomes the limitations of conventional USOD methods, which either implicitly encode physical mechanisms or apply simplistic denoising. Evaluated on the USOD10K benchmark, our method achieves a 0.072 improvement in the structural-measure (Sₘ), surpassing state-of-the-art approaches. This demonstrates the effectiveness and generalizability of synergistically integrating physical priors with temporal dynamics in diffusion-based USOD.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Learning to Search
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Web Mining and Content Analysis: Content-based information diffusion

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 S_m, demonstrating the effectiveness and superiority of our method. The code will be published after the acceptance.

Problem

• Addresses underwater image degradation and domain gaps in salient object detection
• Incorporates underwater physical imaging priors into neural network training
• Enhances salient object identification through temporal dimension modeling

Innovation

• Incorporates underwater physics as explicit training priors
• Uses rectified flow framework for salient object detection
• Introduces temporal dimension modeling to enhance identification