ACMU-nets: Attention cascading modular U-nets incorporating squeeze and excitation blocks

Seokjun Kang; Brian Kenji Iwana; Seiichi Uchida

doi:10.1007/978-3-030-57058-3_9

ACMU-nets: Attention cascading modular U-nets incorporating squeeze and excitation blocks

Seokjun Kang, Brian Kenji Iwana, Seiichi Uchida

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

Abstract

In document analysis research, image-to-image conversion models such as a U-Net have been shown significant performance. Recently, cascaded U-Nets research is suggested for solving complex document analysis studies. However, improving performance by adding U-Net modules requires using too many parameters in cascaded U-Nets. Therefore, in this paper, we propose a method for enhancing the performance of cascaded U-Nets. We suggest a novel document image binarization method by utilizing Cascading Modular U-Nets (CMU-Nets) and Squeeze and Excitation blocks (SE-blocks). Through verification experiments, we point out the problems caused by the use of SE-blocks in existing CMU-Nets and suggest how to use SE-blocks in CMU-Nets. We use the Document Image Binarization (DIBCO) 2017 dataset to evaluate the proposed model.

Original language	English
Title of host publication	Document Analysis Systems - 14th IAPR International Workshop, DAS 2020, Proceedings
Editors	Xiang Bai, Dimosthenis Karatzas, Daniel Lopresti
Publisher	Springer
Pages	118-130
Number of pages	13
ISBN (Print)	9783030570576
DOIs	https://doi.org/10.1007/978-3-030-57058-3_9
Publication status	Published - 2020
Event	14th IAPR International Workshop on Document Analysis Systems, DAS 2020 - Wuhan, China Duration: Jul 26 2020 → Jul 29 2020

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12116 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	14th IAPR International Workshop on Document Analysis Systems, DAS 2020
Country/Territory	China
City	Wuhan
Period	7/26/20 → 7/29/20

All Science Journal Classification (ASJC) codes

Theoretical Computer Science
Computer Science(all)

Access to Document

10.1007/978-3-030-57058-3_9

Cite this

Kang, S., Iwana, B. K., & Uchida, S. (2020). ACMU-nets: Attention cascading modular U-nets incorporating squeeze and excitation blocks. In X. Bai, D. Karatzas, & D. Lopresti (Eds.), Document Analysis Systems - 14th IAPR International Workshop, DAS 2020, Proceedings (pp. 118-130). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12116 LNCS). Springer. https://doi.org/10.1007/978-3-030-57058-3_9

ACMU-nets : Attention cascading modular U-nets incorporating squeeze and excitation blocks. / Kang, Seokjun; Iwana, Brian Kenji ; Uchida, Seiichi.

Document Analysis Systems - 14th IAPR International Workshop, DAS 2020, Proceedings. ed. / Xiang Bai; Dimosthenis Karatzas; Daniel Lopresti. Springer, 2020. p. 118-130 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12116 LNCS).

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

Kang, S, Iwana, BK & Uchida, S 2020, ACMU-nets: Attention cascading modular U-nets incorporating squeeze and excitation blocks. in X Bai, D Karatzas & D Lopresti (eds), Document Analysis Systems - 14th IAPR International Workshop, DAS 2020, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12116 LNCS, Springer, pp. 118-130, 14th IAPR International Workshop on Document Analysis Systems, DAS 2020, Wuhan, China, 7/26/20. https://doi.org/10.1007/978-3-030-57058-3_9

Kang S, Iwana BK , Uchida S. ACMU-nets: Attention cascading modular U-nets incorporating squeeze and excitation blocks. In Bai X, Karatzas D, Lopresti D, editors, Document Analysis Systems - 14th IAPR International Workshop, DAS 2020, Proceedings. Springer. 2020. p. 118-130. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-57058-3_9

Kang, Seokjun ; Iwana, Brian Kenji ; Uchida, Seiichi. / ACMU-nets : Attention cascading modular U-nets incorporating squeeze and excitation blocks. Document Analysis Systems - 14th IAPR International Workshop, DAS 2020, Proceedings. editor / Xiang Bai ; Dimosthenis Karatzas ; Daniel Lopresti. Springer, 2020. pp. 118-130 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{c6e5f11b220249b784626bafc2edb292,

title = "ACMU-nets: Attention cascading modular U-nets incorporating squeeze and excitation blocks",

abstract = "In document analysis research, image-to-image conversion models such as a U-Net have been shown significant performance. Recently, cascaded U-Nets research is suggested for solving complex document analysis studies. However, improving performance by adding U-Net modules requires using too many parameters in cascaded U-Nets. Therefore, in this paper, we propose a method for enhancing the performance of cascaded U-Nets. We suggest a novel document image binarization method by utilizing Cascading Modular U-Nets (CMU-Nets) and Squeeze and Excitation blocks (SE-blocks). Through verification experiments, we point out the problems caused by the use of SE-blocks in existing CMU-Nets and suggest how to use SE-blocks in CMU-Nets. We use the Document Image Binarization (DIBCO) 2017 dataset to evaluate the proposed model.",

author = "Seokjun Kang and Iwana, {Brian Kenji} and Seiichi Uchida",

note = "Funding Information: This work was supported by JSPS KAKENHI Grant Number JP17K19402 and JP17H06100. Publisher Copyright: {\textcopyright} Springer Nature Switzerland AG 2020.; 14th IAPR International Workshop on Document Analysis Systems, DAS 2020 ; Conference date: 26-07-2020 Through 29-07-2020",

year = "2020",

doi = "10.1007/978-3-030-57058-3_9",

language = "English",

isbn = "9783030570576",

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

publisher = "Springer",

pages = "118--130",

editor = "Xiang Bai and Dimosthenis Karatzas and Daniel Lopresti",

booktitle = "Document Analysis Systems - 14th IAPR International Workshop, DAS 2020, Proceedings",

}

TY - GEN

T1 - ACMU-nets

T2 - 14th IAPR International Workshop on Document Analysis Systems, DAS 2020

AU - Kang, Seokjun

AU - Iwana, Brian Kenji

AU - Uchida, Seiichi

N1 - Funding Information: This work was supported by JSPS KAKENHI Grant Number JP17K19402 and JP17H06100. Publisher Copyright: © Springer Nature Switzerland AG 2020.

PY - 2020

Y1 - 2020

N2 - In document analysis research, image-to-image conversion models such as a U-Net have been shown significant performance. Recently, cascaded U-Nets research is suggested for solving complex document analysis studies. However, improving performance by adding U-Net modules requires using too many parameters in cascaded U-Nets. Therefore, in this paper, we propose a method for enhancing the performance of cascaded U-Nets. We suggest a novel document image binarization method by utilizing Cascading Modular U-Nets (CMU-Nets) and Squeeze and Excitation blocks (SE-blocks). Through verification experiments, we point out the problems caused by the use of SE-blocks in existing CMU-Nets and suggest how to use SE-blocks in CMU-Nets. We use the Document Image Binarization (DIBCO) 2017 dataset to evaluate the proposed model.

AB - In document analysis research, image-to-image conversion models such as a U-Net have been shown significant performance. Recently, cascaded U-Nets research is suggested for solving complex document analysis studies. However, improving performance by adding U-Net modules requires using too many parameters in cascaded U-Nets. Therefore, in this paper, we propose a method for enhancing the performance of cascaded U-Nets. We suggest a novel document image binarization method by utilizing Cascading Modular U-Nets (CMU-Nets) and Squeeze and Excitation blocks (SE-blocks). Through verification experiments, we point out the problems caused by the use of SE-blocks in existing CMU-Nets and suggest how to use SE-blocks in CMU-Nets. We use the Document Image Binarization (DIBCO) 2017 dataset to evaluate the proposed model.

UR - http://www.scopus.com/inward/record.url?scp=85090094559&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85090094559&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-57058-3_9

DO - 10.1007/978-3-030-57058-3_9

M3 - Conference contribution

AN - SCOPUS:85090094559

SN - 9783030570576

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

SP - 118

EP - 130

BT - Document Analysis Systems - 14th IAPR International Workshop, DAS 2020, Proceedings

A2 - Bai, Xiang

A2 - Karatzas, Dimosthenis

A2 - Lopresti, Daniel

PB - Springer

Y2 - 26 July 2020 through 29 July 2020

ER -

ACMU-nets: Attention cascading modular U-nets incorporating squeeze and excitation blocks

Abstract

Publication series

Conference

All Science Journal Classification (ASJC) codes

Access to Document

Other files and links

Fingerprint

Cite this