A Source-Free Unsupervised Domain Adaptation Method Based on Feature Consistency

Joon Ho Lee, Gyemin Lee

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

A typical unsupervised domain adaptation (UDA) task assumes that labeled source data is available for model adaptation. However, this assumption is often infeasible due to confidentiality issues or memory constraints on mobile devices. To address these problems, we propose a simple yet effective source-free UDA method. Motivated by self-training methods, the proposed method constrains multiple views of the same image to have similar features and prediction outputs. The feature generator is encouraged to learn consistent visual features that are away from the decision boundaries of the head classifier. On popular benchmarks, our approach demonstrates comparable or even superior performance to vanilla UDA methods without using source images.

Original languageEnglish
Title of host publicationInternational Workshop on Advanced Imaging Technology, IWAIT 2023
EditorsMasayuki Nakajima, Jae-Gon Kim, Kwang-deok Seo, Toshihiko Yamasaki, Jing-Ming Guo, Phooi Yee Lau, Qian Kemao
PublisherSPIE
ISBN (Electronic)9781510663084
DOIs
StatePublished - 2023
Event2023 International Workshop on Advanced Imaging Technology, IWAIT 2023 - Jeju, Korea, Republic of
Duration: 9 Jan 202311 Jan 2023

Publication series

NameProceedings of SPIE - The International Society for Optical Engineering
Volume12592
ISSN (Print)0277-786X
ISSN (Electronic)1996-756X

Conference

Conference2023 International Workshop on Advanced Imaging Technology, IWAIT 2023
Country/TerritoryKorea, Republic of
CityJeju
Period9/01/2311/01/23

Keywords

  • Feature consistency
  • Self-training
  • Source-free unsupervised domain adaptation

Fingerprint

Dive into the research topics of 'A Source-Free Unsupervised Domain Adaptation Method Based on Feature Consistency'. Together they form a unique fingerprint.

Cite this