Learning Image-to-Image Translation Using Paired and Unpaired Training Samples

Soumya Tripathy; Juho Kannala; Esa Rahtu

doi:10.1007/978-3-030-20890-5_4

Learning Image-to-Image Translation Using Paired and Unpaired Training Samples

Soumya Tripathy^*, Juho Kannala, Esa Rahtu

^*Corresponding author for this work

Tampere University

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Scientific › peer-review

8 Citations (Scopus)

Abstract

Image-to-image translation is a general name for a task where an image from one domain is converted to a corresponding image in another domain, given sufficient training data. Traditionally different approaches have been proposed depending on whether aligned image pairs or two sets of (unaligned) examples from both domains are available for training. While paired training samples might be difficult to obtain, the unpaired setup leads to a highly under-constrained problem and inferior results. In this paper, we propose a new general purpose image-to-image translation model that is able to utilize both paired and unpaired training data simultaneously. We compare our method with two strong baselines and obtain both qualitatively and quantitatively improved results. Our model outperforms the baselines also in the case of purely paired and unpaired training data. To our knowledge, this is the first work to consider such hybrid setup in image-to-image translation.

Original language	English
Title of host publication	Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers
Editors	Konrad Schindler, Greg Mori, C.V. Jawahar, Hongdong Li
Publisher	Springer
Pages	51-66
Number of pages	16
ISBN (Print)	9783030208899
DOIs	https://doi.org/10.1007/978-3-030-20890-5_4
Publication status	Published - 2019
MoE publication type	A4 Article in a conference publication
Event	Asian Conference on Computer Vision - Perth, Australia Duration: 2 Dec 2018 → 6 Dec 2018 Conference number: 14

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Publisher	Springer Nature
Volume	11362 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	Asian Conference on Computer Vision
Abbreviated title	ACCV
Country/Territory	Australia
City	Perth
Period	02/12/2018 → 06/12/2018

Access to Document

10.1007/978-3-030-20890-5_4

https://arxiv.org/abs/1805.03189

OpenUrl availability

Full text

Cite this

Tripathy, S., Kannala, J., & Rahtu, E. (2019). Learning Image-to-Image Translation Using Paired and Unpaired Training Samples. In K. Schindler, G. Mori, C. V. Jawahar, & H. Li (Eds.), Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers (pp. 51-66). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11362 LNCS). Springer. https://doi.org/10.1007/978-3-030-20890-5_4

Tripathy, Soumya ; Kannala, Juho ; Rahtu, Esa. / Learning Image-to-Image Translation Using Paired and Unpaired Training Samples. Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers. editor / Konrad Schindler ; Greg Mori ; C.V. Jawahar ; Hongdong Li. Springer, 2019. pp. 51-66 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{4365c31db37d4dbda622b05b05fca87e,

title = "Learning Image-to-Image Translation Using Paired and Unpaired Training Samples",

abstract = "Image-to-image translation is a general name for a task where an image from one domain is converted to a corresponding image in another domain, given sufficient training data. Traditionally different approaches have been proposed depending on whether aligned image pairs or two sets of (unaligned) examples from both domains are available for training. While paired training samples might be difficult to obtain, the unpaired setup leads to a highly under-constrained problem and inferior results. In this paper, we propose a new general purpose image-to-image translation model that is able to utilize both paired and unpaired training data simultaneously. We compare our method with two strong baselines and obtain both qualitatively and quantitatively improved results. Our model outperforms the baselines also in the case of purely paired and unpaired training data. To our knowledge, this is the first work to consider such hybrid setup in image-to-image translation.",

author = "Soumya Tripathy and Juho Kannala and Esa Rahtu",

year = "2019",

doi = "10.1007/978-3-030-20890-5_4",

language = "English",

isbn = "9783030208899",

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

publisher = "Springer",

pages = "51--66",

editor = "Konrad Schindler and Greg Mori and C.V. Jawahar and Hongdong Li",

booktitle = "Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers",

address = "Germany",

note = "Asian Conference on Computer Vision , ACCV ; Conference date: 02-12-2018 Through 06-12-2018",

}

Tripathy, S, Kannala, J & Rahtu, E 2019, Learning Image-to-Image Translation Using Paired and Unpaired Training Samples. in K Schindler, G Mori, CV Jawahar & H Li (eds), Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11362 LNCS, Springer, pp. 51-66, Asian Conference on Computer Vision , Perth, Western Australia, Australia, 02/12/2018. https://doi.org/10.1007/978-3-030-20890-5_4

Learning Image-to-Image Translation Using Paired and Unpaired Training Samples. / Tripathy, Soumya; Kannala, Juho; Rahtu, Esa.
Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers. ed. / Konrad Schindler; Greg Mori; C.V. Jawahar; Hongdong Li. Springer, 2019. p. 51-66 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11362 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Scientific › peer-review

TY - GEN

T1 - Learning Image-to-Image Translation Using Paired and Unpaired Training Samples

AU - Tripathy, Soumya

AU - Kannala, Juho

AU - Rahtu, Esa

N1 - Conference code: 14

PY - 2019

Y1 - 2019

N2 - Image-to-image translation is a general name for a task where an image from one domain is converted to a corresponding image in another domain, given sufficient training data. Traditionally different approaches have been proposed depending on whether aligned image pairs or two sets of (unaligned) examples from both domains are available for training. While paired training samples might be difficult to obtain, the unpaired setup leads to a highly under-constrained problem and inferior results. In this paper, we propose a new general purpose image-to-image translation model that is able to utilize both paired and unpaired training data simultaneously. We compare our method with two strong baselines and obtain both qualitatively and quantitatively improved results. Our model outperforms the baselines also in the case of purely paired and unpaired training data. To our knowledge, this is the first work to consider such hybrid setup in image-to-image translation.

AB - Image-to-image translation is a general name for a task where an image from one domain is converted to a corresponding image in another domain, given sufficient training data. Traditionally different approaches have been proposed depending on whether aligned image pairs or two sets of (unaligned) examples from both domains are available for training. While paired training samples might be difficult to obtain, the unpaired setup leads to a highly under-constrained problem and inferior results. In this paper, we propose a new general purpose image-to-image translation model that is able to utilize both paired and unpaired training data simultaneously. We compare our method with two strong baselines and obtain both qualitatively and quantitatively improved results. Our model outperforms the baselines also in the case of purely paired and unpaired training data. To our knowledge, this is the first work to consider such hybrid setup in image-to-image translation.

UR - http://www.scopus.com/inward/record.url?scp=85067340886&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-20890-5_4

DO - 10.1007/978-3-030-20890-5_4

M3 - Conference contribution

AN - SCOPUS:85067340886

SN - 9783030208899

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

SP - 51

EP - 66

BT - Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers

A2 - Schindler, Konrad

A2 - Mori, Greg

A2 - Jawahar, C.V.

A2 - Li, Hongdong

PB - Springer

T2 - Asian Conference on Computer Vision

Y2 - 2 December 2018 through 6 December 2018

ER -

Tripathy S, Kannala J, Rahtu E. Learning Image-to-Image Translation Using Paired and Unpaired Training Samples. In Schindler K, Mori G, Jawahar CV, Li H, editors, Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers. Springer. 2019. p. 51-66. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-20890-5_4

Learning Image-to-Image Translation Using Paired and Unpaired Training Samples

Abstract

Publication series

Conference

Access to Document

OpenUrl availability

Other files and links

Fingerprint

Cite this