Learning in Compressed Domain for Faster Machine Vision Tasks

Jinming Liu, Heming Sun*, Jiro Katto

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Learned image compression (LIC) has illustrated good ability for reconstruction quality driven tasks (e.g. PSNR, MS-SSIM) and machine vision tasks such as image understanding. However, most LIC frameworks are based on pixel domain, which requires the decoding process. In this paper, we develop a learned compressed domain framework for machine vision tasks. 1) By sending the compressed latent representation directly to the task network, the decoding computation can be eliminated to reduce the complexity. 2) By sorting the latent channels by entropy, only selective channels will be transmitted to the task network, which can reduce the bitrate. As a result, compared with the traditional pixel domain methods, we can reduce about 1/3 multiply-add operations (MACs) and 1/5 inference time while keeping the same accuracy. Moreover, proposed channel selection can contribute to at most 6.8% bitrate saving.

Original languageEnglish
Title of host publication2021 International Conference on Visual Communications and Image Processing, VCIP 2021 - Proceedings
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781728185514
DOIs
Publication statusPublished - 2021
Event2021 International Conference on Visual Communications and Image Processing, VCIP 2021 - Munich, Germany
Duration: 2021 Dec 52021 Dec 8

Publication series

Name2021 International Conference on Visual Communications and Image Processing, VCIP 2021 - Proceedings

Conference

Conference2021 International Conference on Visual Communications and Image Processing, VCIP 2021
Country/TerritoryGermany
CityMunich
Period21/12/521/12/8

Keywords

  • Compressed domain
  • Face alignment
  • Image compression
  • Video coding for machine

ASJC Scopus subject areas

  • Computer Networks and Communications
  • Computer Vision and Pattern Recognition
  • Signal Processing
  • Sensory Systems
  • Communication

Fingerprint

Dive into the research topics of 'Learning in Compressed Domain for Faster Machine Vision Tasks'. Together they form a unique fingerprint.

Cite this