Style-Decoupled Adaptive Routing Network for Underwater Image Enhancement

Hang Xu; Chen Long; Bing Wang; Hao Chen; Zhen Dong

cs.CVApr 14, 2026

Style-Decoupled Adaptive Routing Network for Underwater Image Enhancement

Authors: Hang Xu, Chen Long, Bing Wang, Hao Chen, Zhen Dong

Organizations: State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, 430079, China · Department of Aeronautical and Aviation Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong Special Administrative Region of China

Abstract

Underwater Image Enhancement (UIE) is essential for robust visual perception in marine applications. However, existing methods predominantly rely on uniform mapping tailored to average dataset distributions, leading to over-processing mildly degraded images or insufficient recovery for severe ones. To address this challenge, we propose a novel adaptive enhancement framework, SDAR-Net. Unlike existing uniform paradigms, it first decouples specific degradation styles from the input and subsequently modulates the enhancement process adaptively. Specifically, since underwater degradation primarily shifts the appearance while keeping the scene structure, SDAR-Net formulates image features into dynamic degradation style embeddings and static scene structural representations through a carefully designed training framework. Subsequently, we introduce an adaptive routing mechanism. By evaluating style features and adaptively predicting soft weights at different enhancement states, it guides the weighted fusion of the corresponding image representations, accurately satisfying the adaptive restoration demands of each image. Extensive experiments show that SDAR-Net achieves a new state-of-the-art (SOTA) performance with a PSNR of 25.72 dB on real-world benchmark, and demonstrates its utility in downstream vision tasks. Our code is available at https://github.com/WHU-USI3DV/SDAR-Net.

Figures & tables

Figure 1: (a) Previous methods applying uniform enhancement strategies yield suboptimal results across various degradation degree. (b) Our approach adaptively modulates the enhancement process under the guidance of degradation patterns, addressing the edge-cases and enhancing overall performance.

Figure 2: Overall framework of SDAR-Net. This framework consists of two parts: (a) Representation Decoupling; (b) Adaptive Trajectory Modulation.

Figure 3: Design of lightweight encoder (E) and decoder (D). Conv-in and Conv-out are convolutional layers, Grad denotes the gradient extraction operation and Down denotes downsample operation.

Figure 4: Design of SREU. REB denotes Representation Enhancement Block, SEB denotes Style Evolution Block.

Figure 5: Design of Ada-Route module. T denotes transpose, Proj-A and Proj-B denotes linear projection layer aiming to compress the information from Gram matrix.

Method	UIEB					LSUI
Method	PSNR $\uparrow$	SSIM $\uparrow$	UIQM $\uparrow$	UCIQE $\uparrow$	MUSIQ $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	UIQM $\uparrow$	UCIQE $\uparrow$	MUSIQ $\uparrow$
UDCP	11.0736	0.5153	3.6838	0.6149	59.3019	13.1057	0.5623	4.1156	0.5982	47.6424
Ucolor	19.9437	0.7642	4.1840	0.5879	55.1289	20.1013	0.7003	3.9241	0.5572	35.8873
UWCNN	20.0796	0.8933	4.3106	0.5808	60.5273	23.4756	0.8944	4.3802	0.5716	53.4860
U-shape	22.6897	0.8391	4.3122	0.6205	56.5789	26.9908	0.8765	4.3781	0.5905	51.2496
DPF-Net	23.0872	0.8518	4.2407	0.6056	63.7496	21.8208	0.8345	4.3645	0.6015	49.9639

Table 1: Quantitative comparison on UIEB and LSUI dataset. Higher values are better. Best and second-best results are bolded and underlined , respectively.

Figure 6: Qualitative comparison on UIEB dataset. We mark the output image with the corresponding PSNR.

Figure 7: Enhancement trajectory visualize with routing condition vectors. (a) represents the common cases that can be covered by state 1. (b) represents cases that need a deeper enhancement trajectory. (c) represents cases that only requires a small degree of enhancement. (d) represents that the optimal trajectory is between two candidate states.

w/o Decouple	w/o Ada-Route	PSNR $\uparrow$	MUSIQ $\uparrow$
×	×	25.0219	61.8405
✓	×	25.3101	61.7947
×	✓	25.0065	62.1304
✓	✓	25.7249	61.8258

Table 2: Ablation study on UIEB. We evaluate all combinations of the representation decoupling (Decouple) and the adaptive trajectory modulation (Ada-Route).

$K$	PSNR $\uparrow$	MUSIQ $\uparrow$	$w_{0}$	$w_{1}$	$w_{2}$	$w_{3}$	$w_{4}$
1	25.4759	62.0685	0.0449	0.9551	-	-	-
2	25.7249	61.8258	0.0311	0.8201	0.1488	-	-
3	25.4688	61.3879	0.0414	0.8461	0.1097	0.0028	-
4	25.6128	61.5108	0.0487	0.8874	0.0504	0.0134	0.0001

Table 3: Ablation on the number of enhancement iterations.

Figure 8: Routing effect on mild degradation cases. We mark the output image with the corresponding PSNR and weighted results with its routing weights.

Figure 9: Routing effect on severe degradation cases. We mark the output image with the corresponding PSNR and weighted results with its routing weights.

Strategies	PSNR $\uparrow$	MUSIQ $\uparrow$
State 0	17.3298	59.6402
State 1	25.4238	61.8111
State 2	22.3304	61.5854
External cascade	22.0784	61.7479
Internal cascade	25.4681	61.7842
Weighted	25.7249	61.8258

Table 4: Analysis of routing effectiveness.

Method	Avg	BW	HD	PF	WR	RO	RI	FV	SR
Baseline	0.6046	0.8457	0.6686	0.2107	0.5672	0.6092	0.6049	0.6581	0.6723
Ucolor	0.5780	0.8401	0.6657	0.0196	0.6205	0.6107	0.6121	0.6428	0.6122
U-shape	0.5831	0.8584	0.7130	0.0578	0.6275	0.4654	0.6017	0.6984	0.6428
UDCP	0.5903	0.8512	0.6800	0.0716	0.5686	0.6384	0.5876	0.7066	0.6182
UWCNN	0.6051	0.8540	0.7023	0.2073	0.6142	0.5646	0.6072	0.6458	0.6457
HCLR-Net	0.6125	0.8492	0.6669	0.2344	0.6061	0.6137	0.6309	0.6966	0.6017

Table 5: Quantitative comparison of effectiveness on underwater semantic segmentation. The metric is mIoU, higher values are better. Best and second-best results are bolded and underlined , respectively.

Explore similar work

RPL-UIE: Reliable Prior Learning for Underwater Image Enhancement
Jul 31, 2026Yifan Chen, Jiaming Liu, Ye Zheng +2Diffusion Models for Image Restoration Underwater Image Enhancement
College of Future Information Technology, Fudan University, Shanghai 200433, China · Institute of Artificial Intelligence (TeleAI), China Telecom, China · Faculty of Robot Science and Engineering, Northeastern University, Shenyang 110819, China +1
Real-Time Underwater Image Enhancement via Frequency-Guided Dual-Path Attention
Jun 29, 2026Leshen Zhang, Ao Li, Ce ZhuImage Enhancement Underwater Image Enhancement
Glasgow College, UESTC, Chengdu, China · University of Glasgow, Glasgow, UK · School of Information and Communication Engineering, UESTC, Chengdu, China
CoRe-UIE: Rethinking Coexisting and Region-wise Degradation for Underwater Image Enhancement
Aug 10, 2026Weifeng Kong, Chenghao Xu, Lin Chen +2Underwater Image Enhancement Mixture of Experts
Hohai University Jiangsu, China

Jul 31, 2026J/K move · Enter open · S save

RPL-UIE: Reliable Prior Learning for Underwater Image Enhancement

Yifan Chen, Jiaming Liu, Ye Zheng, Zhe Sun+1

College of Future Information Technology, Fudan University, Shanghai 200433, China · Institute of Artificial Intelligence (TeleAI), China Telecom, China · Faculty of Robot Science and Engineering, Northeastern University, Shenyang 110819, China+1

Diffusion Models for Image Restoration Underwater Image Enhancement

Open paper AskPDF arXiv

Underwater image enhancement (UIE) aims to recover clear images from observations affected by wavelength-dependent absorption, scattering, and spatially nonuniform degradation. Although existing generative methods can handle complex degradations, severe information loss may lead to semantic drift in the restored results. To address this issue, we propose RPL-UIE, a two-stage teacher--student framework for reliable prior learning. In the teacher stage, the network learns reliable and complementary spatial priors characterizing appearance and photometric properties from paired degraded and reference images. In the student stage, the network takes only degraded images as input and learns to emulate the teacher's prior extraction capability, thereby providing more reliable restoration guidance for the enhancement process without requiring reference images at inference. To reduce the prior-learning discrepancy between the teacher and student models, we further develop Residual Prior Refinement Diffusion (RPRD) and Frequency-Aware Prior Residual Calibration (FPRC). RPRD uses the coarse priors as anchors and progressively predicts the necessary corrections in the residual space. FPRC retains stable low-frequency residual components and selectively modulates high-frequency detail residuals, producing calibrated priors to support high-quality reconstruction. Experiments on multiple UIE benchmarks demonstrate competitive restoration performance. Downstream underwater object detection and instance segmentation experiments further demonstrate the improved utility of enhanced images for visual perception, while tests on real-world data captured by a remotely operated vehicle (ROV) support the practical applicability of RPL-UIE.

Abstract

Explore similar work

RPL-UIE: Reliable Prior Learning for Underwater Image Enhancement

Real-Time Underwater Image Enhancement via Frequency-Guided Dual-Path Attention

CoRe-UIE: Rethinking Coexisting and Region-wise Degradation for Underwater Image Enhancement