Learning in Compressed Domain for Faster Machine Vision Tasks

Jinming Liu, Heming Sun*, Jiro Katto

*この研究の対応する著者

研究成果: Conference contribution

抄録

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.

本文言語English
ホスト出版物のタイトル2021 International Conference on Visual Communications and Image Processing, VCIP 2021 - Proceedings
出版社Institute of Electrical and Electronics Engineers Inc.
ISBN(電子版)9781728185514
DOI
出版ステータスPublished - 2021
イベント2021 International Conference on Visual Communications and Image Processing, VCIP 2021 - Munich, Germany
継続期間: 2021 12月 52021 12月 8

出版物シリーズ

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

Conference

Conference2021 International Conference on Visual Communications and Image Processing, VCIP 2021
国/地域Germany
CityMunich
Period21/12/521/12/8

ASJC Scopus subject areas

  • コンピュータ ネットワークおよび通信
  • コンピュータ ビジョンおよびパターン認識
  • 信号処理
  • 感覚系
  • 通信

フィンガープリント

「Learning in Compressed Domain for Faster Machine Vision Tasks」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル