Efficient Scale-Recurrent Network Using Generative Adversarial Network for Image Deblurring

Wei Hsiang Hsu; Chih Wei Tang

doi:10.1109/ICCE53296.2022.9730407

Efficient Scale-Recurrent Network Using Generative Adversarial Network for Image Deblurring

Wei Hsiang Hsu, Chih Wei Tang

Department of Communication Engineering

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

1 Scopus citations

Abstract

Ubiquitous blurry images degrade viewing experiences and performance of video analysis. To be applicable to consumer electronics, the large amount of parameters and high computational load of deblurring network have to be avoided. The existing SRN+ has a small number of parameters (3.9M) and comparable performance. To improve the quality of outputs of a deblurring network (e.g., SRN+) without changing its architecture, this paper proposes the pseudo label based order task for training the discriminator. The proposed funnel soft label further reduces the problem of vanishing gradient during training SRN+ (generator), and the adversarial loss combined with weighted scale-level losses improves quality of deblurring. For GoPro dataset, the proposed scheme outperforms the light version of the state-of-the-art MPRNet in PSNR (+1dB) and number of parameters (70%).

Original language	English
Title of host publication	2022 IEEE International Conference on Consumer Electronics, ICCE 2022
Publisher	Institute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)	9781665441544
DOIs	https://doi.org/10.1109/ICCE53296.2022.9730407
State	Published - 2022
Event	2022 IEEE International Conference on Consumer Electronics, ICCE 2022 - Virtual, Online, United States Duration: 7 Jan 2022 → 9 Jan 2022

Publication series

Name	Digest of Technical Papers - IEEE International Conference on Consumer Electronics
Volume	2022-January
ISSN (Print)	0747-668X

Conference

Conference	2022 IEEE International Conference on Consumer Electronics, ICCE 2022
Country/Territory	United States
City	Virtual, Online
Period	7/01/22 → 9/01/22

Keywords

Image deblurring
generative adversarial networks (GAN)
loss function
pseudo label
scale recurrent network

Access to Document

10.1109/ICCE53296.2022.9730407

Generative Adversarial Network Based Vi Sual Tracking
Tang, C.-W. (PI)
1/08/20 → 31/07/21
Project: Research
Deep Learning Based 360-Degree Vi Deo Processing and Analysis
Tang, C.-W. (PI)
1/08/19 → 31/07/20
Project: Research

Cite this

Hsu, W. H., & Tang, C. W. (2022). Efficient Scale-Recurrent Network Using Generative Adversarial Network for Image Deblurring. In 2022 IEEE International Conference on Consumer Electronics, ICCE 2022 (Digest of Technical Papers - IEEE International Conference on Consumer Electronics; Vol. 2022-January). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICCE53296.2022.9730407

@inproceedings{c154ad09217e440bb54ec06f88b7f40b,

title = "Efficient Scale-Recurrent Network Using Generative Adversarial Network for Image Deblurring",

abstract = "Ubiquitous blurry images degrade viewing experiences and performance of video analysis. To be applicable to consumer electronics, the large amount of parameters and high computational load of deblurring network have to be avoided. The existing SRN+ has a small number of parameters (3.9M) and comparable performance. To improve the quality of outputs of a deblurring network (e.g., SRN+) without changing its architecture, this paper proposes the pseudo label based order task for training the discriminator. The proposed funnel soft label further reduces the problem of vanishing gradient during training SRN+ (generator), and the adversarial loss combined with weighted scale-level losses improves quality of deblurring. For GoPro dataset, the proposed scheme outperforms the light version of the state-of-the-art MPRNet in PSNR (+1dB) and number of parameters (70%).",

keywords = "Image deblurring, generative adversarial networks (GAN), loss function, pseudo label, scale recurrent network",

author = "Hsu, {Wei Hsiang} and Tang, {Chih Wei}",

note = "Publisher Copyright: {\textcopyright} 2022 IEEE.; 2022 IEEE International Conference on Consumer Electronics, ICCE 2022 ; Conference date: 07-01-2022 Through 09-01-2022",

year = "2022",

doi = "10.1109/ICCE53296.2022.9730407",

language = "???core.languages.en_GB???",

series = "Digest of Technical Papers - IEEE International Conference on Consumer Electronics",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

booktitle = "2022 IEEE International Conference on Consumer Electronics, ICCE 2022",

}

Hsu, WH & Tang, CW 2022, Efficient Scale-Recurrent Network Using Generative Adversarial Network for Image Deblurring. in 2022 IEEE International Conference on Consumer Electronics, ICCE 2022. Digest of Technical Papers - IEEE International Conference on Consumer Electronics, vol. 2022-January, Institute of Electrical and Electronics Engineers Inc., 2022 IEEE International Conference on Consumer Electronics, ICCE 2022, Virtual, Online, United States, 7/01/22. https://doi.org/10.1109/ICCE53296.2022.9730407

Efficient Scale-Recurrent Network Using Generative Adversarial Network for Image Deblurring. / Hsu, Wei Hsiang; Tang, Chih Wei.
2022 IEEE International Conference on Consumer Electronics, ICCE 2022. Institute of Electrical and Electronics Engineers Inc., 2022. (Digest of Technical Papers - IEEE International Conference on Consumer Electronics; Vol. 2022-January).

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

TY - GEN

T1 - Efficient Scale-Recurrent Network Using Generative Adversarial Network for Image Deblurring

AU - Hsu, Wei Hsiang

AU - Tang, Chih Wei

PY - 2022

Y1 - 2022

N2 - Ubiquitous blurry images degrade viewing experiences and performance of video analysis. To be applicable to consumer electronics, the large amount of parameters and high computational load of deblurring network have to be avoided. The existing SRN+ has a small number of parameters (3.9M) and comparable performance. To improve the quality of outputs of a deblurring network (e.g., SRN+) without changing its architecture, this paper proposes the pseudo label based order task for training the discriminator. The proposed funnel soft label further reduces the problem of vanishing gradient during training SRN+ (generator), and the adversarial loss combined with weighted scale-level losses improves quality of deblurring. For GoPro dataset, the proposed scheme outperforms the light version of the state-of-the-art MPRNet in PSNR (+1dB) and number of parameters (70%).

AB - Ubiquitous blurry images degrade viewing experiences and performance of video analysis. To be applicable to consumer electronics, the large amount of parameters and high computational load of deblurring network have to be avoided. The existing SRN+ has a small number of parameters (3.9M) and comparable performance. To improve the quality of outputs of a deblurring network (e.g., SRN+) without changing its architecture, this paper proposes the pseudo label based order task for training the discriminator. The proposed funnel soft label further reduces the problem of vanishing gradient during training SRN+ (generator), and the adversarial loss combined with weighted scale-level losses improves quality of deblurring. For GoPro dataset, the proposed scheme outperforms the light version of the state-of-the-art MPRNet in PSNR (+1dB) and number of parameters (70%).

KW - Image deblurring

KW - generative adversarial networks (GAN)

KW - loss function

KW - pseudo label

KW - scale recurrent network

UR - http://www.scopus.com/inward/record.url?scp=85127032749&partnerID=8YFLogxK

U2 - 10.1109/ICCE53296.2022.9730407

DO - 10.1109/ICCE53296.2022.9730407

M3 - 會議論文篇章

AN - SCOPUS:85127032749

T3 - Digest of Technical Papers - IEEE International Conference on Consumer Electronics

BT - 2022 IEEE International Conference on Consumer Electronics, ICCE 2022

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2022 IEEE International Conference on Consumer Electronics, ICCE 2022

Y2 - 7 January 2022 through 9 January 2022

ER -

Efficient Scale-Recurrent Network Using Generative Adversarial Network for Image Deblurring

Abstract

Publication series

Conference

Keywords

Access to Document

Fingerprint

Projects

Generative Adversarial Network Based Vi Sual Tracking

Deep Learning Based 360-Degree Vi Deo Processing and Analysis

Cite this