End-to-End Compression Towards Machine Vision: Network Architecture Design and Optimization

The visual signal compression is a long-standing problem. Fueled by the recent advances of deep learning, exciting progress has been made. Despite better compression performance, existing end-to-end compression algorithms are still designed towards better signal quality in terms of rate-distortion o...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Shurun Wang, Zhao Wang, Shiqi Wang, Yan Ye
Formato:	article
Lenguaje:	EN
Publicado:	IEEE 2021
Materias:	Visual signal compression machine vision object detection rate-distortion optimization Electric apparatus and materials. Electric circuits. Electric networks TK452-454.4
Acceso en línea:	https://doaj.org/article/b43d4dd06e9b42119e0080179cb9d2ce
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:b43d4dd06e9b42119e0080179cb9d2ce
record_format	dspace
spelling	oai:doaj.org-article:b43d4dd06e9b42119e0080179cb9d2ce2021-11-24T00:04:14ZEnd-to-End Compression Towards Machine Vision: Network Architecture Design and Optimization2644-122510.1109/OJCAS.2021.3126061https://doaj.org/article/b43d4dd06e9b42119e0080179cb9d2ce2021-01-01T00:00:00Zhttps://ieeexplore.ieee.org/document/9625654/https://doaj.org/toc/2644-1225The visual signal compression is a long-standing problem. Fueled by the recent advances of deep learning, exciting progress has been made. Despite better compression performance, existing end-to-end compression algorithms are still designed towards better signal quality in terms of rate-distortion optimization. In this paper, we show that the design and optimization of network architecture could be further improved for compression towards machine vision. We propose an inverted bottleneck structure for the encoder of the end-to-end compression towards machine vision, which specifically accounts for efficient representation of the semantic information. Moreover, we quest the capability of optimization by incorporating the analytics accuracy into the optimization process, and the optimality is further explored with generalized rate-accuracy optimization in an iterative manner. We use object detection as a showcase for end-to-end compression towards machine vision, and extensive experiments show that the proposed scheme achieves significant BD-rate savings in terms of analysis performance. Moreover, the promise of the scheme is also demonstrated with strong generalization capability towards other machine vision tasks, due to the enabling of signal-level reconstruction.Shurun WangZhao WangShiqi WangYan YeIEEEarticleVisual signal compressionmachine visionobject detectionrate-distortion optimizationElectric apparatus and materials. Electric circuits. Electric networksTK452-454.4ENIEEE Open Journal of Circuits and Systems, Vol 2, Pp 675-685 (2021)
institution	DOAJ
collection	DOAJ
language	EN
topic	Visual signal compression machine vision object detection rate-distortion optimization Electric apparatus and materials. Electric circuits. Electric networks TK452-454.4
spellingShingle	Visual signal compression machine vision object detection rate-distortion optimization Electric apparatus and materials. Electric circuits. Electric networks TK452-454.4 Shurun Wang Zhao Wang Shiqi Wang Yan Ye End-to-End Compression Towards Machine Vision: Network Architecture Design and Optimization
description	The visual signal compression is a long-standing problem. Fueled by the recent advances of deep learning, exciting progress has been made. Despite better compression performance, existing end-to-end compression algorithms are still designed towards better signal quality in terms of rate-distortion optimization. In this paper, we show that the design and optimization of network architecture could be further improved for compression towards machine vision. We propose an inverted bottleneck structure for the encoder of the end-to-end compression towards machine vision, which specifically accounts for efficient representation of the semantic information. Moreover, we quest the capability of optimization by incorporating the analytics accuracy into the optimization process, and the optimality is further explored with generalized rate-accuracy optimization in an iterative manner. We use object detection as a showcase for end-to-end compression towards machine vision, and extensive experiments show that the proposed scheme achieves significant BD-rate savings in terms of analysis performance. Moreover, the promise of the scheme is also demonstrated with strong generalization capability towards other machine vision tasks, due to the enabling of signal-level reconstruction.
format	article
author	Shurun Wang Zhao Wang Shiqi Wang Yan Ye
author_facet	Shurun Wang Zhao Wang Shiqi Wang Yan Ye
author_sort	Shurun Wang
title	End-to-End Compression Towards Machine Vision: Network Architecture Design and Optimization
title_short	End-to-End Compression Towards Machine Vision: Network Architecture Design and Optimization
title_full	End-to-End Compression Towards Machine Vision: Network Architecture Design and Optimization
title_fullStr	End-to-End Compression Towards Machine Vision: Network Architecture Design and Optimization
title_full_unstemmed	End-to-End Compression Towards Machine Vision: Network Architecture Design and Optimization
title_sort	end-to-end compression towards machine vision: network architecture design and optimization
publisher	IEEE
publishDate	2021
url	https://doaj.org/article/b43d4dd06e9b42119e0080179cb9d2ce
work_keys_str_mv	AT shurunwang endtoendcompressiontowardsmachinevisionnetworkarchitecturedesignandoptimization AT zhaowang endtoendcompressiontowardsmachinevisionnetworkarchitecturedesignandoptimization AT shiqiwang endtoendcompressiontowardsmachinevisionnetworkarchitecturedesignandoptimization AT yanye endtoendcompressiontowardsmachinevisionnetworkarchitecturedesignandoptimization
_version_	1718416116037976064

End-to-End Compression Towards Machine Vision: Network Architecture Design and Optimization

Ejemplares similares