Identity-aware Feature Decoupling Learning for Clothing-change Person Re-identification

2025-01-10

AI Summary

To address the vulnerability of identity features to clothing variations in person re-identification (Re-ID) under clothing-change scenarios, this paper proposes an identity-aware feature disentanglement learning framework. Methodologically, it introduces a dual-branch CNN architecture comprising a backbone stream and an attention stream, where clothing-masked images guide spatial attention to emphasize identity-critical regions (e.g., face and body). A novel clothing deviation suppression module is designed to mitigate semantic inconsistency between the two streams, and an identity attention mechanism is developed to disentangle and enhance identity-specific features across diverse appearances. Extensive experiments on benchmark datasets—including Market-1501 and DukeMTMC-reID—demonstrate substantial improvements over baseline methods: identity feature discriminability and clothing-change robustness are significantly enhanced. The results validate the effectiveness of explicit disentanglement modeling between identity and clothing factors.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Feature Construction/Reformulation
• Cognitive Modeling & Cognitive Systems: Social Cognition And Interaction

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Clothing-change person re-identification (CC Re-ID) has attracted increasing attention in recent years due to its application prospect. Most existing works struggle to adequately extract the ID-related information from the original RGB images. In this paper, we propose an Identity-aware Feature Decoupling (IFD) learning framework to mine identity-related features. Particularly, IFD exploits a dual stream architecture that consists of a main stream and an attention stream. The attention stream takes the clothing-masked images as inputs and derives the identity attention weights for effectively transferring the spatial knowledge to the main stream and highlighting the regions with abundant identity-related information. To eliminate the semantic gap between the inputs of two streams, we propose a clothing bias diminishing module specific to the main stream to regularize the features of clothing-relevant regions. Extensive experimental results demonstrate that our framework outperforms other baseline models on several widely-used CC Re-ID datasets.

Problem

• Clothing Change
• Face Recognition
• Identity Verification

Innovation

• Identity-aware Feature Decoupling
• Attention Mechanism
• Clothing Bias Reduction