Lifelong Person Re-identification via Privacy-Preserving Data Replay

2025-08-03

AI Summary

Lifelong person re-identification (LReID) confronts a fundamental tension between privacy preservation and knowledge accumulation in continual learning: storing raw historical samples risks data leakage, whereas exemplar-free methods suffer from catastrophic forgetting. To address this, we propose Privacy-preserving Replay (Pr²R), a novel framework that distills multi-image features into a single synthetic image in pixel space—enabling lightweight, exemplar-free replay without retaining original data. Pr²R further introduces a dual alignment mechanism—jointly optimizing feature-level and style-level consistency—to concurrently mitigate domain shift and representation forgetting. Our approach integrates knowledge distillation, pixel-wise compression, stylized replay, and domain adaptation. Extensive experiments demonstrate state-of-the-art performance across multiple benchmarks, achieving 4–6% absolute mAP gains on sequential tasks. To our knowledge, Pr²R is the first method to achieve high-accuracy continual re-identification under strict privacy constraints—i.e., without storing any raw historical images.

Technology Category

• Machine Learning: Privacy
• Computer Vision: Bias, Fairness & Privacy
• Search and Optimization: Learning to Search

Application Category

• Security and Privacy: Privacy-enhancing technologies
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems

Abstract

Lifelong person re-identification (LReID) aims to incrementally accumulate knowledge across a sequence of tasks under domain shifts. Recently, replay-based methods have demonstrated strong effectiveness in LReID by rehearsing past samples stored in an auxiliary memory. However, storing historical exemplars raises concerns over data privacy. To avoid this, exemplar-free approaches attempt to match the distribution of past data without storing raw samples. Despite being privacy-friendly, these methods often suffer from performance degradation due to the forgetting of specific past knowledge representations. To this end, we propose to condense information from sequential data into the pixel space in the replay memory, enabling Privacy-Preserving Replay (Pr^2R). More specifically, by distilling the training characteristics of multiple real images into a single image, the condensed samples undergo pixel-level changes. This not only protects the privacy of the original data but also makes the replay samples more representative for sequential tasks. During the style replay phase, we align the current domain to the previous one while simultaneously adapting the replay samples to match the style of the current domain. This dual-alignment strategy effectively mitigates both class-incremental challenges and forgetting caused by domain shifts. Extensive experiments on multiple benchmarks show that the proposed method significantly improves replay effectiveness while preserving data privacy. Specifically, Pr^2R achieves 4% and 6% higher accuracy on sequential tasks compared to the current state-of-the-art and other replay-based methods, respectively.

Problem

• Address privacy concerns in lifelong person re-identification
• Mitigate performance degradation from forgetting past knowledge
• Align domain shifts to improve replay effectiveness

Innovation

• Condenses data into pixel space for privacy
• Aligns current and previous domain styles
• Improves replay effectiveness with dual-alignment