Lossy Compression for Embedded Computer Vision Systems

Li Guo, Dajiang Zhou, Jinjia Zhou, Shinji Kimura, Satoshi Goto

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

Computer vision applications are rapidly gaining popularity in embedded systems, which typically involve a difficult trade-off between vision performance and energy consumption under a constraint of real-time processing throughput. Recently, hardware (FPGA and ASIC-based) implementations have emerged that significantly improve the energy efficiency of vision computation. These implementations, however, often involve intensive memory traffic that retains a significant portion of energy consumption at the system level. To address this issue, we are the first researchers to present a lossy compression framework to exploit the trade-off between vision performance and memory traffic for input images. To meet various requirements for memory access patterns in the vision system, a line-to-block format conversion is designed for the framework. Differential pulse-code modulation-based gradient-oriented quantization is developed as the lossy compression algorithm. We also present its hardware design that supports up to 12-scale 1080p@60fps real-time processing. For histogram of oriented gradient-based deformable part models on VOC2007, the proposed framework achieves a 49.6%–60.5% memory traffic reduction at a detection rate degradation of 0.05%–0.34%. For AlexNet on ImageNet, memory traffic reduction achieves up to 60.8% with less than 0.61% classification rate degradation. Compared with the power consumption reduction from memory traffic, the overhead involved for the proposed input image compression is less than 5%.

Original languageEnglish
JournalIEEE Access
DOIs
Publication statusAccepted/In press - 2018 Jul 3

Fingerprint

Computer vision
Data storage equipment
Energy utilization
Differential pulse code modulation
Hardware
Degradation
Application specific integrated circuits
Processing
Image compression
Embedded systems
Energy efficiency
Field programmable gate arrays (FPGA)
Electric power utilization
Throughput

Keywords

  • Computer vision
  • Computer vision
  • Feature extraction
  • feature extraction
  • Hardware
  • Image coding
  • lossy compression
  • Memory management
  • memory traffic reduction
  • Power demand
  • Random access memory

ASJC Scopus subject areas

  • Computer Science(all)
  • Materials Science(all)
  • Engineering(all)

Cite this

Lossy Compression for Embedded Computer Vision Systems. / Guo, Li; Zhou, Dajiang; Zhou, Jinjia; Kimura, Shinji; Goto, Satoshi.

In: IEEE Access, 03.07.2018.

Research output: Contribution to journalArticle

Guo, Li ; Zhou, Dajiang ; Zhou, Jinjia ; Kimura, Shinji ; Goto, Satoshi. / Lossy Compression for Embedded Computer Vision Systems. In: IEEE Access. 2018.
@article{e3fd3d87acb0469f8e3c19f7fb3b38b2,
title = "Lossy Compression for Embedded Computer Vision Systems",
abstract = "Computer vision applications are rapidly gaining popularity in embedded systems, which typically involve a difficult trade-off between vision performance and energy consumption under a constraint of real-time processing throughput. Recently, hardware (FPGA and ASIC-based) implementations have emerged that significantly improve the energy efficiency of vision computation. These implementations, however, often involve intensive memory traffic that retains a significant portion of energy consumption at the system level. To address this issue, we are the first researchers to present a lossy compression framework to exploit the trade-off between vision performance and memory traffic for input images. To meet various requirements for memory access patterns in the vision system, a line-to-block format conversion is designed for the framework. Differential pulse-code modulation-based gradient-oriented quantization is developed as the lossy compression algorithm. We also present its hardware design that supports up to 12-scale 1080p@60fps real-time processing. For histogram of oriented gradient-based deformable part models on VOC2007, the proposed framework achieves a 49.6{\%}–60.5{\%} memory traffic reduction at a detection rate degradation of 0.05{\%}–0.34{\%}. For AlexNet on ImageNet, memory traffic reduction achieves up to 60.8{\%} with less than 0.61{\%} classification rate degradation. Compared with the power consumption reduction from memory traffic, the overhead involved for the proposed input image compression is less than 5{\%}.",
keywords = "Computer vision, Computer vision, Feature extraction, feature extraction, Hardware, Image coding, lossy compression, Memory management, memory traffic reduction, Power demand, Random access memory",
author = "Li Guo and Dajiang Zhou and Jinjia Zhou and Shinji Kimura and Satoshi Goto",
year = "2018",
month = "7",
day = "3",
doi = "10.1109/ACCESS.2018.2852809",
language = "English",
journal = "IEEE Access",
issn = "2169-3536",
publisher = "Institute of Electrical and Electronics Engineers Inc.",

}

TY - JOUR

T1 - Lossy Compression for Embedded Computer Vision Systems

AU - Guo, Li

AU - Zhou, Dajiang

AU - Zhou, Jinjia

AU - Kimura, Shinji

AU - Goto, Satoshi

PY - 2018/7/3

Y1 - 2018/7/3

N2 - Computer vision applications are rapidly gaining popularity in embedded systems, which typically involve a difficult trade-off between vision performance and energy consumption under a constraint of real-time processing throughput. Recently, hardware (FPGA and ASIC-based) implementations have emerged that significantly improve the energy efficiency of vision computation. These implementations, however, often involve intensive memory traffic that retains a significant portion of energy consumption at the system level. To address this issue, we are the first researchers to present a lossy compression framework to exploit the trade-off between vision performance and memory traffic for input images. To meet various requirements for memory access patterns in the vision system, a line-to-block format conversion is designed for the framework. Differential pulse-code modulation-based gradient-oriented quantization is developed as the lossy compression algorithm. We also present its hardware design that supports up to 12-scale 1080p@60fps real-time processing. For histogram of oriented gradient-based deformable part models on VOC2007, the proposed framework achieves a 49.6%–60.5% memory traffic reduction at a detection rate degradation of 0.05%–0.34%. For AlexNet on ImageNet, memory traffic reduction achieves up to 60.8% with less than 0.61% classification rate degradation. Compared with the power consumption reduction from memory traffic, the overhead involved for the proposed input image compression is less than 5%.

AB - Computer vision applications are rapidly gaining popularity in embedded systems, which typically involve a difficult trade-off between vision performance and energy consumption under a constraint of real-time processing throughput. Recently, hardware (FPGA and ASIC-based) implementations have emerged that significantly improve the energy efficiency of vision computation. These implementations, however, often involve intensive memory traffic that retains a significant portion of energy consumption at the system level. To address this issue, we are the first researchers to present a lossy compression framework to exploit the trade-off between vision performance and memory traffic for input images. To meet various requirements for memory access patterns in the vision system, a line-to-block format conversion is designed for the framework. Differential pulse-code modulation-based gradient-oriented quantization is developed as the lossy compression algorithm. We also present its hardware design that supports up to 12-scale 1080p@60fps real-time processing. For histogram of oriented gradient-based deformable part models on VOC2007, the proposed framework achieves a 49.6%–60.5% memory traffic reduction at a detection rate degradation of 0.05%–0.34%. For AlexNet on ImageNet, memory traffic reduction achieves up to 60.8% with less than 0.61% classification rate degradation. Compared with the power consumption reduction from memory traffic, the overhead involved for the proposed input image compression is less than 5%.

KW - Computer vision

KW - Computer vision

KW - Feature extraction

KW - feature extraction

KW - Hardware

KW - Image coding

KW - lossy compression

KW - Memory management

KW - memory traffic reduction

KW - Power demand

KW - Random access memory

UR - http://www.scopus.com/inward/record.url?scp=85049478697&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85049478697&partnerID=8YFLogxK

U2 - 10.1109/ACCESS.2018.2852809

DO - 10.1109/ACCESS.2018.2852809

M3 - Article

JO - IEEE Access

JF - IEEE Access

SN - 2169-3536

ER -