Semantic Feature Learning for Universal Unsupervised Cross-Domain Retrieval addresses challenges in cross-domain retrieval by proposing a two-stage framework. The study focuses on U2CDR, aiming to retrieve samples with distinct category spaces across domains. By establishing a unified prototypical structure and preserving it during domain alignment, the proposed approach outperforms existing works in various scenarios.
The study emphasizes the importance of accurate supervision in cross-domain retrieval methods and highlights the need for unsupervised techniques. The proposed Unified, Enhanced, and Matched (UEM) semantic feature learning framework tackles challenges like distinguishing data samples without labels and achieving alignment across domains without pairing information.
Through extensive experiments on multiple datasets including Office-31, Office-Home, and DomainNet, the UEM framework demonstrates significant performance improvements over state-of-the-art methods. The results validate the effectiveness of UEM in solving U2CDR challenges comprehensively.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Lixu Wang,Xi... at arxiv.org 03-12-2024
https://arxiv.org/pdf/2403.05690.pdfDeeper Inquiries