Improved Generative Convolution Method for Image Generation

Seung Park; Min Uk Yang; Geun Hyeong Kim; Jueng Eun Im; Ki Hun Kim; Yong Goo Shin

doi:10.1109/ICTC55196.2022.9952596

Improved Generative Convolution Method for Image Generation

Seung Park
, Min Uk Yang
, Geun Hyeong Kim
, Jueng Eun Im
, Ki Hun Kim
, Yong Goo Shin

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

Abstract

In our recent study, we introduced a novel convolution method, called GConv, which improves the performance of generative adversarial networks (GAN) by modulating the convolution kernels following the given latent vector. In this paper, we analyze the limitations of GConv, and propose an improved GConv to address those problems. While GConv modulates the convolution kernel equally at all pixels, the proposed method produces pixel-wise different kernels following not only the given latent vector but also the feature in each pixel. Even though the proposed method is a simple modification of GConv, it shows better performance compared to the standard convolution as well as GConv. To show the superiority of the proposed method, this paper provides experimental results on the CIFAR-10 and CIFAR-100 datasets. Quantitative evaluations reveal that the proposed method improves both GAN and conditional GAN (cGAN) performance in terms of Frechet inception distance (FID).

Original language	English
Title of host publication	ICTC 2022 - 13th International Conference on Information and Communication Technology Convergence
Subtitle of host publication	Accelerating Digital Transformation with ICT Innovation
Publisher	IEEE Computer Society
Pages	546-551
Number of pages	6
ISBN (Electronic)	9781665499392
DOIs	https://doi.org/10.1109/ICTC55196.2022.9952596
Publication status	Published - 2022
Externally published	Yes
Event	13th International Conference on Information and Communication Technology Convergence, ICTC 2022 - Jeju Island, Korea, Republic of Duration: 2022 Oct 19 → 2022 Oct 21

Publication series

Name	International Conference on ICT Convergence
Volume	2022-October
ISSN (Print)	2162-1233
ISSN (Electronic)	2162-1241

Conference

Conference	13th International Conference on Information and Communication Technology Convergence, ICTC 2022
Country/Territory	Korea, Republic of
City	Jeju Island
Period	22/10/19 → 22/10/21

Bibliographical note

Publisher Copyright:
© 2022 IEEE.

Keywords

Convolution operation
GConv
Generative adversarial networks
Image generation

ASJC Scopus subject areas

Information Systems
Computer Networks and Communications

Access to Document

10.1109/ICTC55196.2022.9952596

Cite this

Park, S., Yang, M. U., Kim, G. H., Im, J. E., Kim, K. H., & Shin, Y. G. (2022). Improved Generative Convolution Method for Image Generation. In ICTC 2022 - 13th International Conference on Information and Communication Technology Convergence: Accelerating Digital Transformation with ICT Innovation (pp. 546-551). (International Conference on ICT Convergence; Vol. 2022-October). IEEE Computer Society. https://doi.org/10.1109/ICTC55196.2022.9952596

Improved Generative Convolution Method for Image Generation. / Park, Seung; Yang, Min Uk; Kim, Geun Hyeong et al.
ICTC 2022 - 13th International Conference on Information and Communication Technology Convergence: Accelerating Digital Transformation with ICT Innovation. IEEE Computer Society, 2022. p. 546-551 (International Conference on ICT Convergence; Vol. 2022-October).

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

Park, S, Yang, MU, Kim, GH, Im, JE, Kim, KH & Shin, YG 2022, Improved Generative Convolution Method for Image Generation. in ICTC 2022 - 13th International Conference on Information and Communication Technology Convergence: Accelerating Digital Transformation with ICT Innovation. International Conference on ICT Convergence, vol. 2022-October, IEEE Computer Society, pp. 546-551, 13th International Conference on Information and Communication Technology Convergence, ICTC 2022, Jeju Island, Korea, Republic of, 22/10/19. https://doi.org/10.1109/ICTC55196.2022.9952596

Park S, Yang MU, Kim GH, Im JE, Kim KH, Shin YG. Improved Generative Convolution Method for Image Generation. In ICTC 2022 - 13th International Conference on Information and Communication Technology Convergence: Accelerating Digital Transformation with ICT Innovation. IEEE Computer Society. 2022. p. 546-551. (International Conference on ICT Convergence). doi: 10.1109/ICTC55196.2022.9952596

@inproceedings{b5c57ff4bc2843a2ae8682cfed69f040,

title = "Improved Generative Convolution Method for Image Generation",

abstract = "In our recent study, we introduced a novel convolution method, called GConv, which improves the performance of generative adversarial networks (GAN) by modulating the convolution kernels following the given latent vector. In this paper, we analyze the limitations of GConv, and propose an improved GConv to address those problems. While GConv modulates the convolution kernel equally at all pixels, the proposed method produces pixel-wise different kernels following not only the given latent vector but also the feature in each pixel. Even though the proposed method is a simple modification of GConv, it shows better performance compared to the standard convolution as well as GConv. To show the superiority of the proposed method, this paper provides experimental results on the CIFAR-10 and CIFAR-100 datasets. Quantitative evaluations reveal that the proposed method improves both GAN and conditional GAN (cGAN) performance in terms of Frechet inception distance (FID).",

keywords = "Convolution operation, GConv, Generative adversarial networks, Image generation",

author = "Seung Park and Yang, \{Min Uk\} and Kim, \{Geun Hyeong\} and Im, \{Jueng Eun\} and Kim, \{Ki Hun\} and Shin, \{Yong Goo\}",

note = "Publisher Copyright: {\textcopyright} 2022 IEEE.; 13th International Conference on Information and Communication Technology Convergence, ICTC 2022 ; Conference date: 19-10-2022 Through 21-10-2022",

year = "2022",

doi = "10.1109/ICTC55196.2022.9952596",

language = "English",

series = "International Conference on ICT Convergence",

publisher = "IEEE Computer Society",

pages = "546--551",

booktitle = "ICTC 2022 - 13th International Conference on Information and Communication Technology Convergence",

}

TY - GEN

T1 - Improved Generative Convolution Method for Image Generation

AU - Park, Seung

AU - Yang, Min Uk

AU - Kim, Geun Hyeong

AU - Im, Jueng Eun

AU - Kim, Ki Hun

AU - Shin, Yong Goo

PY - 2022

Y1 - 2022

N2 - In our recent study, we introduced a novel convolution method, called GConv, which improves the performance of generative adversarial networks (GAN) by modulating the convolution kernels following the given latent vector. In this paper, we analyze the limitations of GConv, and propose an improved GConv to address those problems. While GConv modulates the convolution kernel equally at all pixels, the proposed method produces pixel-wise different kernels following not only the given latent vector but also the feature in each pixel. Even though the proposed method is a simple modification of GConv, it shows better performance compared to the standard convolution as well as GConv. To show the superiority of the proposed method, this paper provides experimental results on the CIFAR-10 and CIFAR-100 datasets. Quantitative evaluations reveal that the proposed method improves both GAN and conditional GAN (cGAN) performance in terms of Frechet inception distance (FID).

AB - In our recent study, we introduced a novel convolution method, called GConv, which improves the performance of generative adversarial networks (GAN) by modulating the convolution kernels following the given latent vector. In this paper, we analyze the limitations of GConv, and propose an improved GConv to address those problems. While GConv modulates the convolution kernel equally at all pixels, the proposed method produces pixel-wise different kernels following not only the given latent vector but also the feature in each pixel. Even though the proposed method is a simple modification of GConv, it shows better performance compared to the standard convolution as well as GConv. To show the superiority of the proposed method, this paper provides experimental results on the CIFAR-10 and CIFAR-100 datasets. Quantitative evaluations reveal that the proposed method improves both GAN and conditional GAN (cGAN) performance in terms of Frechet inception distance (FID).

KW - Convolution operation

KW - GConv

KW - Generative adversarial networks

KW - Image generation

UR - https://www.scopus.com/pages/publications/85143251122

U2 - 10.1109/ICTC55196.2022.9952596

DO - 10.1109/ICTC55196.2022.9952596

M3 - Conference contribution

AN - SCOPUS:85143251122

T3 - International Conference on ICT Convergence

SP - 546

EP - 551

BT - ICTC 2022 - 13th International Conference on Information and Communication Technology Convergence

PB - IEEE Computer Society

T2 - 13th International Conference on Information and Communication Technology Convergence, ICTC 2022

Y2 - 19 October 2022 through 21 October 2022

ER -

Improved Generative Convolution Method for Image Generation

Abstract

Publication series

Conference

Bibliographical note

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this