ChanMin Park's BlogChanMin Park's Blog
  • Dec 07, 2025

[PaperReview] Correspondence-free Domain Alignment for Unsupervised Cross-domain Image Retrieval

 

Wang, Xu, et al. “Correspondence-free domain alignment for unsupervised cross-domain image retrieval.” Proceedings of the AAAI Conference on Artificial Intelligence. Vol. 37. No. 8. 2023.

Paper Link: arXiv:2302.06081

Introduction

This paper presents a correspondence-free domain alignment method for unsupervised cross-domain image retrieval. The approach eliminates the need for paired data between domains by learning domain-invariant representations.

Key Contributions

  • Correspondence-free learning framework that does not require paired samples
  • Domain alignment strategy for cross-domain retrieval
  • Effective feature learning without explicit domain correspondences

Methodology

The proposed method learns domain-invariant features through alignment techniques that do not rely on correspondence information between domains.

Figures

Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6
Figure 7
Figure 8
Figure 9
Figure 10
Figure 11

Results

The method demonstrates improved performance in cross-domain image retrieval tasks compared to existing approaches.