RDM-IR: Task-adaptive deep unfolding network for All-In-One image restoration

Cheng, Y; Shao, M; Wan, Y; Wang, C

RDM-IR: Task-adaptive deep unfolding network for All-In-One image restoration

Cheng, Y Shao, M Wan, Y Wang, C

Permalink

Publisher:: Elsevier
Publication Type:: Journal Article
Citation:: Knowledge-Based Systems, 2024, 304, pp. 112543
Issue Date:: 2024-11-25

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

The embargo period expires on 25 Nov 2025

Adobe PDF

Download Accepted versionAdobe PDF (7.59 MB)

View on publisher's site

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Cheng, Y
dc.contributor.author	Shao, M
dc.contributor.author	Wan, Y
dc.contributor.author	Wang, C https://orcid.org/0000-0003-1297-768X
dc.date.accessioned	2025-03-02T23:58:53Z
dc.date.available	2025-03-02T23:58:53Z
dc.date.issued	2024-11-25
dc.identifier.citation	Knowledge-Based Systems, 2024, 304, pp. 112543
dc.identifier.issn	0950-7051
dc.identifier.uri	http://hdl.handle.net/10453/185479
dc.description.abstract	Existing All-In-One image restoration (IR) methods have made promising progress in unified restoration from different degradations. Nevertheless, these methods lack precise modeling of task-specific degradations, which limits the flexible processing of divergent features in different degradations and leads to suboptimal restoration performance. To address the aforementioned issue, we propose a task-adaptive All-In-One IR framework called RDM-IR, to flexibly tackle exclusive features of different degradations via modeling specific degradations. Specifically, our methods consist of two subtasks: (1) Reference-based Degradation Modeling and (2) model-based IR. The RDM-IR first dynamically models the task-specific degradation based on a referenced image pair and further restores the target image with the collected degradation statistics. Besides, to model the task-specific degradation more precisely, we further devise a Degradation Prior Transmitter (DPT) to introduce degradation information for the corresponding task. DPT embeds the degradation prior from the reference image into the degradation modeling process and simultaneously prevents the interference of non-degradation-related information. The proposed RDM-IR presents superior flexibility in task-specific degradation modeling and reaches state-of-the-art on several benchmark datasets. Our code is open-sourced in the following repository: https://github.com/YuanshuoCheng/RDM-IR.
dc.language	en
dc.publisher	Elsevier
dc.relation.ispartof	Knowledge-Based Systems
dc.relation.isbasedon	10.1016/j.knosys.2024.112543
dc.rights	info:eu-repo/semantics/embargoedAccess
dc.subject	08 Information and Computing Sciences, 15 Commerce, Management, Tourism and Services, 17 Psychology and Cognitive Sciences
dc.subject.classification	Artificial Intelligence & Image Processing
dc.subject.classification	4602 Artificial intelligence
dc.subject.classification	4605 Data management and data science
dc.subject.classification	4611 Machine learning
dc.title	RDM-IR: Task-adaptive deep unfolding network for All-In-One image restoration
dc.type	Journal Article
utslib.citation.volume	304
utslib.for	08 Information and Computing Sciences
utslib.for	15 Commerce, Management, Tourism and Services
utslib.for	17 Psychology and Cognitive Sciences
pubs.organisational-group	University of Technology Sydney
pubs.organisational-group	University of Technology Sydney/Faculty of Engineering and Information Technology
pubs.organisational-group	University of Technology Sydney/Faculty of Engineering and Information Technology/School of Computer Science
utslib.copyright.status	open_access	*
utslib.copyright.embargo	2025-11-25T00:00:00+1000Z
dc.date.updated	2025-03-02T23:58:52Z
pubs.publication-status	Published
pubs.volume	304

Abstract:

Existing All-In-One image restoration (IR) methods have made promising progress in unified restoration from different degradations. Nevertheless, these methods lack precise modeling of task-specific degradations, which limits the flexible processing of divergent features in different degradations and leads to suboptimal restoration performance. To address the aforementioned issue, we propose a task-adaptive All-In-One IR framework called RDM-IR, to flexibly tackle exclusive features of different degradations via modeling specific degradations. Specifically, our methods consist of two subtasks: (1) Reference-based Degradation Modeling and (2) model-based IR. The RDM-IR first dynamically models the task-specific degradation based on a referenced image pair and further restores the target image with the collected degradation statistics. Besides, to model the task-specific degradation more precisely, we further devise a Degradation Prior Transmitter (DPT) to introduce degradation information for the corresponding task. DPT embeds the degradation prior from the reference image into the degradation modeling process and simultaneously prevents the interference of non-degradation-related information. The proposed RDM-IR presents superior flexibility in task-specific degradation modeling and reaches state-of-the-art on several benchmark datasets. Our code is open-sourced in the following repository: https://github.com/YuanshuoCheng/RDM-IR.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/185479