DPC-Net: Dual-Prior Collaborative Network for All-in-One Image Restoration

Zhaokun He; Kangbiao Shi; Axi Niu; Jian Jin; Peng Wu; Wei Dong; Qingsen Yan

cs.CVAug 20, 2026

DPC-Net: Dual-Prior Collaborative Network for All-in-One Image Restoration

Authors: Zhaokun He, Kangbiao Shi, Axi Niu, Jian Jin, Peng Wu, Wei Dong, Qingsen Yan

Organizations: Northwestern Polytechnical University · Singapore Management University · Xi’an University of Architecture and Technology

Abstract

All-in-One Image Restoration (AiOIR) aims to handle diverse degradations within a unified model. However, existing methods often overlook image semantics in degradation modeling and lack low-level visual priors during reconstruction, leading to structural distortions and semantic inconsistencies. To address these issues, we propose a novel Dual-Prior Collaborative Network (DPC-Net), which achieves high-quality restoration by jointly exploiting degradation-semantic coupled priors and low-level visual priors. Specifically, degraded images are fed into a Degradation-Aware Network (DAN) to extract degradation-semantic coupled features. To this end, a Vision-Language Model (VLM) supervises DAN by constraining its features distribution, introducing image semantics into the encoding of degradation patterns. A Degradation-Semantic Modulation Module (DSMM) further translates this guidance into degradation-semantic coupling and propagates coupled representations to the decoder. During decoding, knowledge bases provide low-level visual priors, and the Dual-Prior Collaborative Reconstruction Module (DPCR) integrates dual-prior information to guide degradation removal while preserving structure and semantics, producing high-fidelity restored images. Extensive experiments on multiple restoration benchmarks demonstrate that DPC-Net achieves superior performance against state-of-the-art AiOIR methods.

Figures & tables

Figure 1: Motivations of our method. (a) Models that implicitly learn degradation features. (b) Models that explicitly learn degradation features. (c) Our method.

Figure 2: Overview of our proposed Dual-Prior Collaborative Network (DPC-Net) for all-in-one image restoration.

Figure 3: Illustration of our proposed Dual-Prior Collaborative Reconstruction Module (DPCR).

Figure 4: Comparison of feature maps across different channels.

Method	Source	Params.	Dehazing		Deraining		Denoising						Average
Method	Source	Params.	SOTS		Rain100L		BSD68 $\sigma$ =15		BSD68 $\sigma$ =25		BSD68 $\sigma$ =50		Average
AirNet ( Li et al. 2022 )	CVPR’22	9M	27.94	.962	34.90	.967	33.92	.933	31.26	.888	28.00	.797	31.20	.910
IDR ( Zhang et al. 2023 )	CVPR’23	15M	29.87	.970	36.03	.971	33.89	.931	31.32	.884	28.04	.798	31.83	.911
PromptIR ( Potlapalli et al. 2023 )	NeurIPS’23	33M	30.58	.974	36.37	.972	33.98	.933	31.31	.888	28.06	.799	32.06	.913
NDR ( Yao et al. 2024 )	TIP’24	28M	28.64	.962	35.42	.969	34.01	.932	31.36	.887	28.10	.798	31.51	.910
Gridformer ( Wang et al. 2024 )	IJCV’24	34M	30.37	.970	37.15	.972	33.93	.931	31.37	.887	28.11	.801	32.19	.912

Table 1: Comparison to state-of-the-art all-in-one methods on the three degradation tasks. Best and second best performances are highlighted. PSNR (dB,

\uparrow

) and SSIM (

\uparrow

) metrics are reported on the full RGB images.

Method	Source	Params.	Dehazing		Deraining		Denoising		Deblurring		Low-Light		Average
Method	Source	Params.	SOTS		Rain100L		BSD68 $\sigma$ =25		GoPro		LOLv1		Average
AirNet ( Li et al. 2022 )	CVPR’22	9M	21.04	.884	32.98	.951	30.91	.882	24.35	.781	18.18	.735	25.49	.846
IDR ( Zhang et al. 2023 )	CVPR’23	15M	25.24	.943	35.63	.965	31.60	.887	27.87	.846	21.34	.826	28.34	.893
PromptIR ( Potlapalli et al. 2023 )	NeurIPS’23	33M	26.54	.949	36.37	.970	31.47	.886	28.71	.881	22.68	.832	29.15	.904
Gridformer ( Wang et al. 2024 )	IJCV’24	34M	26.79	.951	36.61	.971	31.45	.885	29.22	.884	22.59	.831	29.33	.904
InstructIR ( Conde et al. 2024 )	ECCV’24	16M	27.10	.956	36.84	.973	31.40	.887	29.40	.886	23.00	.836	29.55	.907

Table 2: Comparison to state-of-the-art all-in-one methods on the five degradation tasks. Best and second best performances are highlighted. PSNR (dB,

\uparrow

) and SSIM (

\uparrow

) metrics are reported on the full RGB images.

Panel of Figure 5 — Figure 5: Visual comparisons of DPC-Net with state-of-the-art all-in-one methods on the three degradation tasks.

Method	Dehazing	Method	Deraining	Method	Denoising
Method	SOTS	Method	Rain100L	Method	BSD68 $\sigma$ =25
MSCNN ( Ren et al. 2016 )	22.06/.908	UMR ( Yasarla and Patel 2019 )	32.39/.921	CBM3D ( Dabov et al. 2007 )	30.69/.868
AODNet ( Li et al. 2017 )	20.29/.877	SIRR ( Wei et al. 2019 )	32.37/.926	DnCNN ( Zhang et al. 2017a )	31.23/.883
EPDN ( Qu et al. 2019 )	22.57/.863	MSPFN ( Jiang et al. 2020 )	33.50/.948	IRCNN ( Zhang et al. 2017b )	31.18/.882
FDGAN ( Dong et al. 2020 )	23.15/.921	LPNet ( Gao et al. 2019 )	23.15/.921	BRDNet ( Tian et al. 2020 )	31.43/.885
AirNet ( Li et al. 2022 )	23.18/.900	AirNet ( Li et al. 2022 )	34.90/.977	AirNet ( Li et al. 2022 )	31.48/.893

Table 3: Comparison to state-of-the-art all-in-one methods on the single degradation task. Best and second best performances are highlighted. PSNR (dB,

\uparrow

) and SSIM (

\uparrow

) metrics are reported on the full RGB images.

Panel of Figure 6 — Figure 6: Visual comparisons of DPC-Net with its variants to demonstrate the progressive improvement brought by each component.

Index	(1)	(2)	(3)	(4)	PSNR $\uparrow$	SSIM $\uparrow$
(a)	✗	✗	✓	✓	32.65	0.920
(b)	✓	✗	✓	✓	32.86	0.920
(c)	✓	✓	✗	✗	32.66	0.919
(d)	✓	✓	✓	✗	32.79	0.920
Ours	✓	✓	✓	✓	33.01	0.922

Table 4: Effectiveness of key components under the three degradation tasks. (1), (2), (3), and (4) denote DAN, DEM, knowledge base, and DPCR, respectively.

Panel of Figure 7 — Figure 7: t-SNE ( Van der Maaten and Hinton 2008 ) visualization of $\mathbf{E}^{n}$ under the three-task setting. Variant (a) corresponds to Tab. 4 .

Index	Method	PSNR $\uparrow$	SSIM $\uparrow$
(a)	Prompt	32.46	0.918
(b)	DA-CLIP	32.64	0.919
(c)	LLaVA (Ours)	33.01	0.922

Table 5: Effectiveness of LLaVA-based guidance strategy under the three degradation tasks.

$m$	w/o QGM		w QGM
$m$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$
128	32.62	0.918	32.73	0.920
256	32.75	0.920	33.01	0.922
512	32.64	0.920	32.81	0.920

Table 6: Effectiveness of the QGM and comparison of knowledge base capacities across three degradation tasks.

Geometry-Guided Image Restoration

See all 16 papers

What to Remove, What to Preserve: Dual-Ambiguity Rectification for All-in-One Image Restoration
Jul 30, 2026Cencen Liu, Wen Yin, Dongyang Zhang +6Representation Learning Representation Disentanglement
University of Electronic Science and Technology of China · University of Electronic Science and Technology of China, Chengdu, Sichuan, China · Jiigan Technology +1
ClusIR: Towards Cluster-Guided All-in-One Image Restoration
Dec 11, 2025Shengkai Hu, Jiaqi Ma, Xu Zhang +3Mixture of Experts All-in-One Image Restoration
Mohamed bin Zayed University of Artificial Intelligence Abu Dhabi, United Arab Emirates · Nanyang Technological University Singapore, Singapore
DVANet: Degradation-aware Visual-prior Alignment Network for Image Restoration
Jun 17, 2026Yanjie Tu, Qingsen Yan, Axi Niu +3All-in-One Image Restoration Image Restoration
School of Computer Science, Northwestern Polytechnical University, Xi’an 710000, China · Shenzhen Research Institute of Northwestern Polytechnical University, Shenzhen 518057, China · State Key Laboratory of Internet of Things for Smart City, University of Macau, Macau 999078, China

Jul 30, 2026J/K move · Enter open · S save

What to Remove, What to Preserve: Dual-Ambiguity Rectification for All-in-One Image Restoration

Cencen Liu, Wen Yin, Dongyang Zhang, Dongmin Li+5

University of Electronic Science and Technology of China · University of Electronic Science and Technology of China, Chengdu, Sichuan, China · Jiigan Technology+1

Representation Learning Representation Disentanglement All-in-One Image Restoration Image Restoration Degradation-Aware Image Restoration

Open paper AskPDF arXiv

All-in-one image restoration aims to handle diverse degradations within a unified framework. Existing methods commonly encode heterogeneous degradation conditions in a shared latent space, where degradation-related cues and scene content can remain entangled. We characterize the resulting challenge as dual ambiguity: semantic ambiguity in channel-wise modulation and spatial ambiguity in restoration responses, which can lead to content corruption and residual artifacts. To mitigate this issue, we propose DAR-Net, a Dual-Ambiguity Rectification Network for all-in-one image restoration. DAR-Net first introduces a Degradation Archetype Representation (DAR) module to construct a structured degradation state through simplex-constrained archetype mixture modeling. Based on this state, a Semantic Ambiguity Rectification (SeAR) module generates degradation-aware prompts to improve channel-wise conditioning in the decoder. A Spatial Ambiguity Rectification (SpAR) module further regularizes degradation-aware and complementary features toward orthogonal response subspaces, reducing spatial interference between removal and preservation cues. Extensive experiments on standard all-in-one restoration benchmarks show that DAR-Net achieves the best overall performance under both three-degradation and five-degradation settings, improving the average PSNR over the strongest competitor by 0.14 dB and 0.34 dB, respectively; it additionally shows superior performance on CDD-11 and WeatherBench.

Abstract

Geometry-Guided Image Restoration

Explore similar work

What to Remove, What to Preserve: Dual-Ambiguity Rectification for All-in-One Image Restoration

ClusIR: Towards Cluster-Guided All-in-One Image Restoration

DVANet: Degradation-aware Visual-prior Alignment Network for Image Restoration