Sublabel-Accurate Multilabeling Meets Product Label Spaces

Zhenzhang Ye, Bjoern Haefner, Yvain Quéau, Thomas Möllenhoff, Daniel Cremers

Publikation: Beitrag in Buch/Bericht/KonferenzbandKonferenzbeitragBegutachtung

Abstract

Functional lifting methods are a promising approach to determine optimal or near-optimal solutions to difficult nonconvex variational problems. Yet, they come with increased memory demands, limiting their practicability. To overcome this drawback, this paper presents a combination of two approaches designed to make liftings more scalable, namely product-space relaxations and sublabel-accurate discretizations. Our main contribution is a simple way to solve the resulting semi-infinite optimization problem with a sampling strategy. We show that despite its simplicity, our approach significantly outperforms baseline methods, in the sense that it finds solutions with lower energies given the same amount of memory. We demonstrate our empirical findings on the nonconvex optical flow and manifold-valued denoising problems.

OriginalspracheEnglisch
TitelPattern Recognition - 43rd DAGM German Conference, DAGM GCPR 2021, Proceedings
Redakteure/-innenChristian Bauckhage, Juergen Gall, Alexander Schwing
Herausgeber (Verlag)Springer Science and Business Media Deutschland GmbH
Seiten3-17
Seitenumfang15
ISBN (Print)9783030926588
DOIs
PublikationsstatusVeröffentlicht - 2021
Veranstaltung43rd DAGM German Conference on Pattern Recognition, DAGM GCPR 2021 - Virtual, Online
Dauer: 28 Sept. 20211 Okt. 2021

Publikationsreihe

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Band13024 LNCS
ISSN (Print)0302-9743
ISSN (elektronisch)1611-3349

Konferenz

Konferenz43rd DAGM German Conference on Pattern Recognition, DAGM GCPR 2021
OrtVirtual, Online
Zeitraum28/09/211/10/21

Fingerprint

Untersuchen Sie die Forschungsthemen von „Sublabel-Accurate Multilabeling Meets Product Label Spaces“. Zusammen bilden sie einen einzigartigen Fingerprint.

Dieses zitieren