End-to-end Image Compression with Swin-Transformer

Meng WANG; Kai ZHANG; Li ZHANG; Yue LI; Junru LI; Yue WANG; Shiqi WANG

doi:10.1109/VCIP56404.2022.10008895

End-to-end Image Compression with Swin-Transformer

Meng WANG^*, Kai ZHANG, Li ZHANG, Yue LI, Junru LI, Yue WANG, Shiqi WANG^*

^*Corresponding author for this work

Research output: Book Chapters | Papers in Conference Proceedings › Conference paper (refereed) › Research › peer-review

3 Citations (Scopus)

Abstract

In this paper, we propose an end-to-end image compression framework, which cooperates with the swin-transformer modules to capture the localized and non-localized similarities in image compression. In particular, the swin-transformer modules are deployed in the analysis and synthesis stages, interleaving with convolution layers. The transformer layers are expected to perceive more flexible receptive fields, such that the spatially localized and non-localized redundancies could be more effectively eliminated. The proposed method reveals the excellent capability of signal conjunction and prediction, leading to the improvement of the rate and distortion performance. Experimental results show that the proposed method is superior to the existing methods on both natural scene and screen content images, where 22.46% BD-Rate savings are achieved when compared with the BPG. Over 30% BD-Rate gains could be observed with screen content images when compared with the classical hyper-prior end-to-end coding method.

Original language	English
Title of host publication	2022 IEEE International Conference on Visual Communications and Image Processing, VCIP 2022
Publisher	IEEE
ISBN (Electronic)	9781665475921
DOIs	https://doi.org/10.1109/VCIP56404.2022.10008895
Publication status	Published - 2022
Externally published	Yes
Event	2022 IEEE International Conference on Visual Communications and Image Processing, VCIP 2022 - Suzhou, China Duration: 13 Dec 2022 → 16 Dec 2022

Conference

Conference	2022 IEEE International Conference on Visual Communications and Image Processing, VCIP 2022
Country/Territory	China
City	Suzhou
Period	13/12/22 → 16/12/22

Bibliographical note

Publisher Copyright:
© 2022 IEEE.

Funding

ACKNOWLEDGMENT This work is supported in part by the National Natural Science Foundation of China under 62022002, in part by the Hong Kong Research Grants Council General Research Fund (GRF) under Grant 11203220, in part by the Hong Kong Innovation and Technology Fund under Grant PRP/059/20FX.

Keywords

convolution
end-to-end compression
Image compression
transformer

Access to Document

10.1109/VCIP56404.2022.10008895

Cite this

@inproceedings{4d364e187d9748e19b663aa814f5be1e,

title = "End-to-end Image Compression with Swin-Transformer",

abstract = "In this paper, we propose an end-to-end image compression framework, which cooperates with the swin-transformer modules to capture the localized and non-localized similarities in image compression. In particular, the swin-transformer modules are deployed in the analysis and synthesis stages, interleaving with convolution layers. The transformer layers are expected to perceive more flexible receptive fields, such that the spatially localized and non-localized redundancies could be more effectively eliminated. The proposed method reveals the excellent capability of signal conjunction and prediction, leading to the improvement of the rate and distortion performance. Experimental results show that the proposed method is superior to the existing methods on both natural scene and screen content images, where 22.46% BD-Rate savings are achieved when compared with the BPG. Over 30% BD-Rate gains could be observed with screen content images when compared with the classical hyper-prior end-to-end coding method.",

keywords = "convolution, end-to-end compression, Image compression, transformer",

author = "Meng WANG and Kai ZHANG and Li ZHANG and Yue LI and Junru LI and Yue WANG and Shiqi WANG",

note = "Publisher Copyright: {\textcopyright} 2022 IEEE.; 2022 IEEE International Conference on Visual Communications and Image Processing, VCIP 2022 ; Conference date: 13-12-2022 Through 16-12-2022",

year = "2022",

doi = "10.1109/VCIP56404.2022.10008895",

language = "English",

booktitle = "2022 IEEE International Conference on Visual Communications and Image Processing, VCIP 2022",

publisher = "IEEE",

}

WANG, M, ZHANG, K, ZHANG, L, LI, Y, LI, J, WANG, Y & WANG, S 2022, End-to-end Image Compression with Swin-Transformer. in 2022 IEEE International Conference on Visual Communications and Image Processing, VCIP 2022. IEEE, 2022 IEEE International Conference on Visual Communications and Image Processing, VCIP 2022, Suzhou, China, 13/12/22. https://doi.org/10.1109/VCIP56404.2022.10008895

TY - GEN

T1 - End-to-end Image Compression with Swin-Transformer

AU - WANG, Meng

AU - ZHANG, Kai

AU - ZHANG, Li

AU - LI, Yue

AU - LI, Junru

AU - WANG, Yue

AU - WANG, Shiqi

PY - 2022

Y1 - 2022

N2 - In this paper, we propose an end-to-end image compression framework, which cooperates with the swin-transformer modules to capture the localized and non-localized similarities in image compression. In particular, the swin-transformer modules are deployed in the analysis and synthesis stages, interleaving with convolution layers. The transformer layers are expected to perceive more flexible receptive fields, such that the spatially localized and non-localized redundancies could be more effectively eliminated. The proposed method reveals the excellent capability of signal conjunction and prediction, leading to the improvement of the rate and distortion performance. Experimental results show that the proposed method is superior to the existing methods on both natural scene and screen content images, where 22.46% BD-Rate savings are achieved when compared with the BPG. Over 30% BD-Rate gains could be observed with screen content images when compared with the classical hyper-prior end-to-end coding method.

AB - In this paper, we propose an end-to-end image compression framework, which cooperates with the swin-transformer modules to capture the localized and non-localized similarities in image compression. In particular, the swin-transformer modules are deployed in the analysis and synthesis stages, interleaving with convolution layers. The transformer layers are expected to perceive more flexible receptive fields, such that the spatially localized and non-localized redundancies could be more effectively eliminated. The proposed method reveals the excellent capability of signal conjunction and prediction, leading to the improvement of the rate and distortion performance. Experimental results show that the proposed method is superior to the existing methods on both natural scene and screen content images, where 22.46% BD-Rate savings are achieved when compared with the BPG. Over 30% BD-Rate gains could be observed with screen content images when compared with the classical hyper-prior end-to-end coding method.

KW - convolution

KW - end-to-end compression

KW - Image compression

KW - transformer

UR - http://www.scopus.com/inward/record.url?scp=85147253833&partnerID=8YFLogxK

U2 - 10.1109/VCIP56404.2022.10008895

DO - 10.1109/VCIP56404.2022.10008895

M3 - Conference paper (refereed)

AN - SCOPUS:85147253833

BT - 2022 IEEE International Conference on Visual Communications and Image Processing, VCIP 2022

PB - IEEE

T2 - 2022 IEEE International Conference on Visual Communications and Image Processing, VCIP 2022

Y2 - 13 December 2022 through 16 December 2022

ER -

End-to-end Image Compression with Swin-Transformer

Abstract

Conference

Bibliographical note

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this