Afficher la notice abrégée

Gradient Step Denoiser for convergent Plug-and-Play

hal.structure.identifierInstitut de Mathématiques de Bordeaux [IMB]
dc.contributor.authorHURAULT, Samuel
hal.structure.identifierInstitut de Mathématiques de Bordeaux [IMB]
dc.contributor.authorLECLAIRE, Arthur
hal.structure.identifierInstitut de Mathématiques de Bordeaux [IMB]
dc.contributor.authorPAPADAKIS, Nicolas
dc.date.accessioned2024-04-04T02:45:12Z
dc.date.available2024-04-04T02:45:12Z
dc.date.conference2022-04-25
dc.identifier.urihttps://oskar-bordeaux.fr/handle/20.500.12278/191448
dc.description.abstractEnPlug-and-Play methods constitute a class of iterative algorithms for imaging problems where regularization is performed by an off-the-shelf denoiser. Although Plug-and-Play methods can lead to tremendous visual performance for various image problems, the few existing convergence guarantees are based on unrealistic (or suboptimal) hypotheses on the denoiser, or limited to strongly convex data terms. In this work, we propose a new type of Plug-and-Play methods, based on half-quadratic splitting, for which the denoiser is realized as a gradient descent step on a functional parameterized by a deep neural network. Exploiting convergence results for proximal gradient descent algorithms in the non-convex setting, we show that the proposed Plug-and-Play algorithm is a convergent iterative scheme that targets stationary points of an explicit global functional. Besides, experiments show that it is possible to learn such a deep denoiser while not compromising the performance in comparison to other state-of-the-art deep denoisers used in Plug-and-Play schemes. We apply our proximal gradient algorithm to various ill-posed inverse problems, e.g. deblurring, super-resolution and inpainting. For all these applications, numerical results empirically confirm the convergence results. Experiments also show that this new algorithm reaches state-of-the-art performance, both quantitatively and qualitatively.
dc.description.sponsorshipRepenser la post-production d'archives avec des méthodes à patch, variationnelles et par apprentissage - ANR-19-CE23-0027
dc.language.isoen
dc.title.enGradient Step Denoiser for convergent Plug-and-Play
dc.typeCommunication dans un congrès
dc.subject.halInformatique [cs]/Traitement des images
dc.subject.halMathématiques [math]/Optimisation et contrôle [math.OC]
dc.subject.halInformatique [cs]/Réseau de neurones [cs.NE]
dc.identifier.arxiv2110.03220
bordeaux.hal.laboratoriesInstitut de Mathématiques de Bordeaux (IMB) - UMR 5251*
bordeaux.institutionUniversité de Bordeaux
bordeaux.institutionBordeaux INP
bordeaux.institutionCNRS
bordeaux.conference.titleInternational Conference on Learning Representations (ICLR'22)
bordeaux.countryUS
bordeaux.conference.cityOnline
bordeaux.peerReviewedoui
hal.identifierhal-03370886
hal.version1
hal.invitednon
hal.proceedingsoui
hal.popularnon
hal.audienceInternationale
hal.origin.linkhttps://hal.archives-ouvertes.fr//hal-03370886v1
bordeaux.COinSctx_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.au=HURAULT,%20Samuel&LECLAIRE,%20Arthur&PAPADAKIS,%20Nicolas&rft.genre=unknown


Fichier(s) constituant ce document

FichiersTailleFormatVue

Il n'y a pas de fichiers associés à ce document.

Ce document figure dans la(les) collection(s) suivante(s)

Afficher la notice abrégée