Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Semantic Feature Learning for Universal Unsupervised Cross-Domain Retrieval

Lixu Wang, Xinyu Du, Qi Zhu

TL;DR

This work introduces the problem of Universal Unsupervised Cross-Domain Retrieval (U^2CDR) for the first time and design a two-stage semantic feature learning framework to address it, demonstrating that this approach significantly outperforms existing state-of-the-art CDR works and some potentially effective studies from other topics in solving U^2CDR challenges.

Abstract

Cross-domain retrieval (CDR), as a crucial tool for numerous technologies, is finding increasingly broad applications. However, existing efforts face several major issues, with the most critical being the need for accurate supervision, which often demands costly resources and efforts. Cutting-edge studies focus on achieving unsupervised CDR but typically assume that the category spaces across domains are identical, an assumption that is often unrealistic in real-world scenarios. This is because only through dedicated and comprehensive analysis can the category spaces of different domains be confirmed as identical, which contradicts the premise of unsupervised scenarios. Therefore, in this work, we introduce the problem of Universal Unsupervised Cross-Domain Retrieval (U^2CDR) for the first time and design a two-stage semantic feature learning framework to address it. In the first stage, a cross-domain unified prototypical structure is established under the guidance of an instance-prototype-mixed contrastive loss and a semantic-enhanced loss, to counteract category space differences. In the second stage, through a modified adversarial training mechanism, we ensure minimal changes for the established prototypical structure during domain alignment, enabling more accurate nearest-neighbor searching. Extensive experiments across multiple datasets and scenarios, including closet, partial, and open-set CDR, demonstrate that our approach significantly outperforms existing state-of-the-art CDR works and some potentially effective studies from other topics in solving U^2CDR challenges.

Semantic Feature Learning for Universal Unsupervised Cross-Domain Retrieval

TL;DR

This work introduces the problem of Universal Unsupervised Cross-Domain Retrieval (U^2CDR) for the first time and design a two-stage semantic feature learning framework to address it, demonstrating that this approach significantly outperforms existing state-of-the-art CDR works and some potentially effective studies from other topics in solving U^2CDR challenges.

Abstract

Cross-domain retrieval (CDR), as a crucial tool for numerous technologies, is finding increasingly broad applications. However, existing efforts face several major issues, with the most critical being the need for accurate supervision, which often demands costly resources and efforts. Cutting-edge studies focus on achieving unsupervised CDR but typically assume that the category spaces across domains are identical, an assumption that is often unrealistic in real-world scenarios. This is because only through dedicated and comprehensive analysis can the category spaces of different domains be confirmed as identical, which contradicts the premise of unsupervised scenarios. Therefore, in this work, we introduce the problem of Universal Unsupervised Cross-Domain Retrieval (U^2CDR) for the first time and design a two-stage semantic feature learning framework to address it. In the first stage, a cross-domain unified prototypical structure is established under the guidance of an instance-prototype-mixed contrastive loss and a semantic-enhanced loss, to counteract category space differences. In the second stage, through a modified adversarial training mechanism, we ensure minimal changes for the established prototypical structure during domain alignment, enabling more accurate nearest-neighbor searching. Extensive experiments across multiple datasets and scenarios, including closet, partial, and open-set CDR, demonstrate that our approach significantly outperforms existing state-of-the-art CDR works and some potentially effective studies from other topics in solving U^2CDR challenges.
Paper Structure (14 sections, 19 equations, 4 figures, 6 tables)

This paper contains 14 sections, 19 equations, 4 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Problem Formulation
  5. Intra-Domain Semantic-Enhanced Learning
  6. Instance-Prototype-Mixed Contrastive Learning.
  7. Semantic-Enhanced Loss.
  8. Cross-Domain Semantic-Matched Learning
  9. Semantic-Preserving Domain Alignment.
  10. Switchable Nearest Neighboring Match.
  11. Experiments
  12. Effectiveness of UEM When Solving U$^2$CDR
  13. Ablation Study
  14. Conclusion

Figures (4)

  • Figure 1: Overview of our proposed UEM semantic feature learning framework for U$^2$CDR. In the first stage of Intra-Domain Semantic-Enhanced Learning, UEM establishes a unified prototypical structure across domains, which is driven and enhanced by IPM and SEL, respectively. Then, in the second stage of Cross-Domain Semantic-Matched Learning, SPDA is used to align domains while preserving the built prototypical structure, and SN$^2$M can achieve more accurate cross-domain categorical pairing.
  • Figure 2: Comparison of nearest neighbor searching before or after domain alignment.
  • Figure 3: Comparison between standard adversarial learning and our semantic-preserving domain alignment in terms of semantic structure changes.
  • Figure 4: Retrieval results of our framework on Office-31 (A$\rightarrow$D) and Office-Home (A$\rightarrow$C) in Closet Unsupervised Cross-Domain Retrieval.