A Dual-stage Prompt-driven Privacy-preserving Paradigm for Person Re-Identification

2025-11-07

AI Summary

To address challenges in person re-identification—including complex synthetic data construction, weak cross-domain generalization, and privacy leakage risks—this paper proposes a two-stage prompt-driven privacy-preserving paradigm. First, it constructs GenePerson, a large-scale anonymized synthetic dataset, leveraging diffusion models and text inversion networks. Second, it introduces a prompt-driven disentanglement mechanism that employs multi-attribute textual prompts to explicitly separate style from content representations, thereby enhancing domain-invariant feature learning. The method integrates prompt engineering, contrastive learning, and an end-to-end generative framework. Evaluated on cross-domain Re-ID benchmarks, models trained on GenePerson with the proposed Prompt-Driven Disentanglement Mechanism (PDM) significantly outperform existing methods relying on either real or synthetic data, achieving state-of-the-art generalization performance.

Technology Category

• Machine Learning: Privacy
• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Prompt Engineering / Prompting

Application Category

• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Security and Privacy: Privacy-enhancing technologies
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

With growing concerns over data privacy, researchers have started using virtual data as an alternative to sensitive real-world images for training person re-identification (Re-ID) models. However, existing virtual datasets produced by game engines still face challenges such as complex construction and poor domain generalization, making them difficult to apply in real scenarios. To address these challenges, we propose a Dual-stage Prompt-driven Privacy-preserving Paradigm (DPPP). In the first stage, we generate rich prompts incorporating multi-dimensional attributes such as pedestrian appearance, illumination, and viewpoint that drive the diffusion model to synthesize diverse data end-to-end, building a large-scale virtual dataset named GenePerson with 130,519 images of 6,641 identities. In the second stage, we propose a Prompt-driven Disentanglement Mechanism (PDM) to learn domain-invariant generalization features. With the aid of contrastive learning, we employ two textual inversion networks to map images into pseudo-words representing style and content, respectively, thereby constructing style-disentangled content prompts to guide the model in learning domain-invariant content features at the image level. Experiments demonstrate that models trained on GenePerson with PDM achieve state-of-the-art generalization performance, surpassing those on popular real and virtual Re-ID datasets.

Problem

• Addresses virtual dataset construction complexity in person re-identification
• Overcomes poor domain generalization of synthetic training data
• Enhances privacy by replacing sensitive real-world images with virtual data

Innovation

• Generates virtual dataset using multi-dimensional prompts
• Employs contrastive learning with textual inversion networks
• Learns domain-invariant features through prompt-driven disentanglement