DM-Adapter: Domain-Aware Mixture-of-Adapters for Text-Based Person Retrieval

• DOI: 10.1609/aaai.v39i6.32608
• Publication Date: 2025-04-11T11:17:25Z
• Authors: Yating Liu; Zimo Liu; Xiangyuan Lan; Wenming Yang; Yaowei Li; Qingmin Liao
• Keywords: Adapter (computing)

Text-based person retrieval (TPR) has gained significant attention as a fine-grained and challenging task that closely aligns with practical applications. Tailoring CLIP to domain is now emerging research topic due the abundant knowledge of vision-language pretraining, but challenges still remain during fine-tuning: (i) Previous full-model fine-tuning in TPR computationally expensive prone overfitting.(ii) Existing parameter-efficient transfer learning (PETL) for lacks feature extraction. To address these issues, we propose Domain-Aware Mixture-of-Adapters (DM-Adapter), which unifies Mixture-of-Experts (MOE) PETL enhance representations while maintaining efficiency. Specifically, Sparse designed parallel MLP layers both vision language branches, where different experts specialize distinct aspects handle features more finely. promote router exploit information effectively alleviate routing imbalance, Router then developed by building novel gating function injecting learnable domain-aware prompts. Extensive experiments show our DM-Adapter achieves state-of-the-art performance, outperforming previous methods margin.