Single-Loss Multi-task Learning For Improving Semantic Segmentation Using Super-Resolution

Andreas Aakerberg; Anders S. Johansen; Kamal Nasrollahi; Thomas B. Moeslund

doi:10.1007/978-3-030-89131-2_37

Single-Loss Multi-task Learning For Improving Semantic Segmentation Using Super-Resolution

Andreas Aakerberg^*, Anders S. Johansen, Kamal Nasrollahi, Thomas B. Moeslund

^*Kontaktforfatter

Publikation: Bidrag til bog/antologi/rapport/konference proceeding › Konferenceartikel i proceeding › Forskning › peer review

2 Citationer (Scopus)

Abstract

We propose a novel means to improve the accuracy of semantic segmentation based on multi-task learning. More specifically, in our Multi-Task Semantic Segmentation and Super-Resolution (MT-SSSR) framework, we jointly train a super-resolution and semantic segmentation model in an end-to-end manner using the same task loss for both models. This allows us to optimize the super-resolution model towards producing images that are optimal for the segmentation task, rather than ones that are of high-fidelity. Simultaneously we adapt the segmentation model to better utilize the improved images and thereby improve the segmentation accuracy. We evaluate our approach on multiple public benchmark datasets, and our extensive experimental results show that our novel MT-SSSR framework outperforms other state-of-the-art approaches.

Originalsprog	Engelsk
Titel	Computer Analysis of Images and Patterns - 19th International Conference, CAIP 2021, Proceedings
Redaktører	Nicolas Tsapatsoulis, Andreas Panayides, Theo Theocharides, Andreas Lanitis, Andreas Lanitis, Constantinos Pattichis, Constantinos Pattichis, Mario Vento
Antal sider	9
Vol/bind	13053
Forlag	Springer Science+Business Media
Publikationsdato	2021
Sider	403-411
ISBN (Trykt)	9783030891305
DOI	https://doi.org/10.1007/978-3-030-89131-2_37
Status	Udgivet - 2021
Begivenhed	19th International Conference on Computer Analysis of Images and Patterns, CAIP 2021 - Virtual, Online Varighed: 28 sep. 2021 → 30 sep. 2021

Konference

Konference	19th International Conference on Computer Analysis of Images and Patterns, CAIP 2021
By	Virtual, Online
Periode	28/09/2021 → 30/09/2021

Navn	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Vol/bind	13053 LNCS
ISSN	0302-9743

Bibliografisk note

Funding Information:
This work was partially supported by the Milestone Research Programme at Aalborg University (MRPA) and Danmarks Frie Forskningsfond (DFF 8022-00360B).

Publisher Copyright:
© 2021, Springer Nature Switzerland AG.

Adgang til dokumentet

10.1007/978-3-030-89131-2_37

AUB Link

Søg efter materialet i Aalborg Universitetsbiblioteks søgemaskine

Andre filer og links

Link to publication in Scopus

Citationsformater

Aakerberg, A., Johansen, A. S., Nasrollahi, K., & Moeslund, T. B. (2021). Single-Loss Multi-task Learning For Improving Semantic Segmentation Using Super-Resolution. I N. Tsapatsoulis, A. Panayides, T. Theocharides, A. Lanitis, A. Lanitis, C. Pattichis, C. Pattichis, & M. Vento (red.), Computer Analysis of Images and Patterns - 19th International Conference, CAIP 2021, Proceedings (Bind 13053, s. 403-411). Springer Science+Business Media. https://doi.org/10.1007/978-3-030-89131-2_37

Aakerberg, Andreas ; Johansen, Anders S. ; Nasrollahi, Kamal et al. / Single-Loss Multi-task Learning For Improving Semantic Segmentation Using Super-Resolution. Computer Analysis of Images and Patterns - 19th International Conference, CAIP 2021, Proceedings. red. / Nicolas Tsapatsoulis ; Andreas Panayides ; Theo Theocharides ; Andreas Lanitis ; Andreas Lanitis ; Constantinos Pattichis ; Constantinos Pattichis ; Mario Vento. Bind 13053 Springer Science+Business Media, 2021. s. 403-411 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Bind 13053 LNCS).

@inproceedings{b8149acdae34466e9bdf502215eafa5f,

title = "Single-Loss Multi-task Learning For Improving Semantic Segmentation Using Super-Resolution",

abstract = "We propose a novel means to improve the accuracy of semantic segmentation based on multi-task learning. More specifically, in our Multi-Task Semantic Segmentation and Super-Resolution (MT-SSSR) framework, we jointly train a super-resolution and semantic segmentation model in an end-to-end manner using the same task loss for both models. This allows us to optimize the super-resolution model towards producing images that are optimal for the segmentation task, rather than ones that are of high-fidelity. Simultaneously we adapt the segmentation model to better utilize the improved images and thereby improve the segmentation accuracy. We evaluate our approach on multiple public benchmark datasets, and our extensive experimental results show that our novel MT-SSSR framework outperforms other state-of-the-art approaches.",

keywords = "Multi-task learning, Semantic segmentation, Super-resolution",

author = "Andreas Aakerberg and Johansen, {Anders S.} and Kamal Nasrollahi and Moeslund, {Thomas B.}",

note = "Publisher Copyright: {\textcopyright} 2021, Springer Nature Switzerland AG.; 19th International Conference on Computer Analysis of Images and Patterns, CAIP 2021 ; Conference date: 28-09-2021 Through 30-09-2021",

year = "2021",

doi = "10.1007/978-3-030-89131-2_37",

language = "English",

isbn = "9783030891305",

volume = "13053",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Science+Business Media",

pages = "403--411",

editor = "Nicolas Tsapatsoulis and Andreas Panayides and Theo Theocharides and Andreas Lanitis and Andreas Lanitis and Constantinos Pattichis and Constantinos Pattichis and Mario Vento",

booktitle = "Computer Analysis of Images and Patterns - 19th International Conference, CAIP 2021, Proceedings",

address = "United States",

}

Aakerberg, A , Johansen, AS , Nasrollahi, K & Moeslund, TB 2021, Single-Loss Multi-task Learning For Improving Semantic Segmentation Using Super-Resolution. i N Tsapatsoulis, A Panayides, T Theocharides, A Lanitis, A Lanitis, C Pattichis, C Pattichis & M Vento (red), Computer Analysis of Images and Patterns - 19th International Conference, CAIP 2021, Proceedings. bind 13053, Springer Science+Business Media, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), bind 13053 LNCS, s. 403-411, 19th International Conference on Computer Analysis of Images and Patterns, CAIP 2021, Virtual, Online, 28/09/2021. https://doi.org/10.1007/978-3-030-89131-2_37

Single-Loss Multi-task Learning For Improving Semantic Segmentation Using Super-Resolution. / Aakerberg, Andreas ; Johansen, Anders S.; Nasrollahi, Kamal et al.
Computer Analysis of Images and Patterns - 19th International Conference, CAIP 2021, Proceedings. red. / Nicolas Tsapatsoulis; Andreas Panayides; Theo Theocharides; Andreas Lanitis; Andreas Lanitis; Constantinos Pattichis; Constantinos Pattichis; Mario Vento. Bind 13053 Springer Science+Business Media, 2021. s. 403-411 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Bind 13053 LNCS).

Publikation: Bidrag til bog/antologi/rapport/konference proceeding › Konferenceartikel i proceeding › Forskning › peer review

TY - GEN

T1 - Single-Loss Multi-task Learning For Improving Semantic Segmentation Using Super-Resolution

AU - Aakerberg, Andreas

AU - Johansen, Anders S.

AU - Nasrollahi, Kamal

AU - Moeslund, Thomas B.

PY - 2021

Y1 - 2021

N2 - We propose a novel means to improve the accuracy of semantic segmentation based on multi-task learning. More specifically, in our Multi-Task Semantic Segmentation and Super-Resolution (MT-SSSR) framework, we jointly train a super-resolution and semantic segmentation model in an end-to-end manner using the same task loss for both models. This allows us to optimize the super-resolution model towards producing images that are optimal for the segmentation task, rather than ones that are of high-fidelity. Simultaneously we adapt the segmentation model to better utilize the improved images and thereby improve the segmentation accuracy. We evaluate our approach on multiple public benchmark datasets, and our extensive experimental results show that our novel MT-SSSR framework outperforms other state-of-the-art approaches.

AB - We propose a novel means to improve the accuracy of semantic segmentation based on multi-task learning. More specifically, in our Multi-Task Semantic Segmentation and Super-Resolution (MT-SSSR) framework, we jointly train a super-resolution and semantic segmentation model in an end-to-end manner using the same task loss for both models. This allows us to optimize the super-resolution model towards producing images that are optimal for the segmentation task, rather than ones that are of high-fidelity. Simultaneously we adapt the segmentation model to better utilize the improved images and thereby improve the segmentation accuracy. We evaluate our approach on multiple public benchmark datasets, and our extensive experimental results show that our novel MT-SSSR framework outperforms other state-of-the-art approaches.

KW - Multi-task learning

KW - Semantic segmentation

KW - Super-resolution

UR - http://www.scopus.com/inward/record.url?scp=85119478818&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-89131-2_37

DO - 10.1007/978-3-030-89131-2_37

M3 - Article in proceeding

AN - SCOPUS:85119478818

SN - 9783030891305

VL - 13053

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 403

EP - 411

BT - Computer Analysis of Images and Patterns - 19th International Conference, CAIP 2021, Proceedings

A2 - Tsapatsoulis, Nicolas

A2 - Panayides, Andreas

A2 - Theocharides, Theo

A2 - Lanitis, Andreas

A2 - Pattichis, Constantinos

A2 - Vento, Mario

PB - Springer Science+Business Media

T2 - 19th International Conference on Computer Analysis of Images and Patterns, CAIP 2021

Y2 - 28 September 2021 through 30 September 2021

ER -

Aakerberg A , Johansen AS , Nasrollahi K , Moeslund TB. Single-Loss Multi-task Learning For Improving Semantic Segmentation Using Super-Resolution. I Tsapatsoulis N, Panayides A, Theocharides T, Lanitis A, Lanitis A, Pattichis C, Pattichis C, Vento M, red., Computer Analysis of Images and Patterns - 19th International Conference, CAIP 2021, Proceedings. Bind 13053. Springer Science+Business Media. 2021. s. 403-411. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Bind 13053 LNCS). doi: 10.1007/978-3-030-89131-2_37

Single-Loss Multi-task Learning For Improving Semantic Segmentation Using Super-Resolution

Abstract

Konference

Bibliografisk note

Adgang til dokumentet

AUB Link

Andre filer og links

Fingeraftryk

Citationsformater