Sublabel-accurate multilabeling meets product label spaces

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

doi:10.1007/978-3-030-92659-5_1

Sublabel-accurate multilabeling meets product label spaces

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

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

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.

Original language	English
Title of host publication	Proceedings of the 43rd DAGM German Conference on Pattern Recognition
Editors	Christian Bauckhage, Juergen Gall, Alexander Schwing
Place of Publication	Cham, Switzerland
Publisher	Springer Cham
Pages	3-17
Number of pages	15
ISBN (Electronic)	9783030926595
ISBN (Print)	9783030926588
DOIs	https://doi.org/10.1007/978-3-030-92659-5_1
Publication status	Published - 13 Jan 2022
Event	43rd DAGM German Conference on Pattern Recognition - Virtual Duration: 28 Sept 2021 → 1 Oct 2021 Conference number: 43 http://www.wikicfp.com/cfp/servlet/event.showcfp?eventid=125183&copyownerid=6365

Publication series

Name	Lecture Notes in Computer Science
Publisher	Springer Cham
Volume	13024
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Symposium

Symposium	43rd DAGM German Conference on Pattern Recognition
Abbreviated title	DAGM GCPR 2021
Period	28/09/21 → 1/10/21
Internet address	http://www.wikicfp.com/cfp/servlet/event.showcfp?eventid=125183&copyownerid=6365

Keywords / Materials (for Non-textual outputs)

variational methods
manifold-valued problems
convex relaxation
global optimization

Access to Document

10.1007/978-3-030-92659-5_1

Cite this

@inproceedings{e23d4de633304a7f8c0d93a367568132,

title = "Sublabel-accurate multilabeling meets product label spaces",

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.",

keywords = "variational methods, manifold-valued problems, convex relaxation, global optimization",

author = "Zhenzhang Ye and Bjoern Haefner and Yvain Qu{\'e}au and Thomas M{\"o}llenhoff and Daniel Cremers",

year = "2022",

month = jan,

day = "13",

doi = "10.1007/978-3-030-92659-5\_1",

language = "English",

isbn = "9783030926588",

series = "Lecture Notes in Computer Science",

publisher = "Springer Cham",

pages = "3--17",

editor = "Christian Bauckhage and Juergen Gall and Alexander Schwing",

booktitle = "Proceedings of the 43rd DAGM German Conference on Pattern Recognition",

note = "43rd DAGM German Conference on Pattern Recognition, DAGM GCPR 2021 ; Conference date: 28-09-2021 Through 01-10-2021",

url = "http://www.wikicfp.com/cfp/servlet/event.showcfp?eventid=125183\&copyownerid=6365",

}

Ye, Z, Haefner, B, Quéau, Y, Möllenhoff, T & Cremers, D 2022, Sublabel-accurate multilabeling meets product label spaces. in C Bauckhage, J Gall & A Schwing (eds), Proceedings of the 43rd DAGM German Conference on Pattern Recognition. Lecture Notes in Computer Science, vol. 13024, Springer Cham, Cham, Switzerland, pp. 3-17, 43rd DAGM German Conference on Pattern Recognition, 28/09/21. https://doi.org/10.1007/978-3-030-92659-5_1

Sublabel-accurate multilabeling meets product label spaces. / Ye, Zhenzhang; Haefner, Bjoern; Quéau, Yvain et al.
Proceedings of the 43rd DAGM German Conference on Pattern Recognition. ed. / Christian Bauckhage; Juergen Gall; Alexander Schwing. Cham, Switzerland: Springer Cham, 2022. p. 3-17 (Lecture Notes in Computer Science; Vol. 13024).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Sublabel-accurate multilabeling meets product label spaces

AU - Ye, Zhenzhang

AU - Haefner, Bjoern

AU - Quéau, Yvain

AU - Möllenhoff, Thomas

AU - Cremers, Daniel

N1 - Conference code: 43

PY - 2022/1/13

Y1 - 2022/1/13

N2 - 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.

AB - 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.

KW - variational methods

KW - manifold-valued problems

KW - convex relaxation

KW - global optimization

U2 - 10.1007/978-3-030-92659-5_1

DO - 10.1007/978-3-030-92659-5_1

M3 - Conference contribution

SN - 9783030926588

T3 - Lecture Notes in Computer Science

SP - 3

EP - 17

BT - Proceedings of the 43rd DAGM German Conference on Pattern Recognition

A2 - Bauckhage, Christian

A2 - Gall, Juergen

A2 - Schwing, Alexander

PB - Springer Cham

CY - Cham, Switzerland

T2 - 43rd DAGM German Conference on Pattern Recognition

Y2 - 28 September 2021 through 1 October 2021

ER -

Sublabel-accurate multilabeling meets product label spaces

Abstract

Publication series

Symposium

Keywords / Materials (for Non-textual outputs)

Access to Document

Fingerprint

Cite this