Multimodal LLM-Empowered Re-Ranking for Generalizable Person Re-Identification

2026-06-14

AI Summary

This work addresses the performance degradation of conventional re-ranking methods in domain-generalized person re-identification (DG Re-ID) under unseen scenarios, which stems from insufficient generalization capability of the underlying encoder. To overcome this limitation, the study introduces a multimodal large language model (MLLM) into the re-ranking stage for the first time. The MLLM is adapted to the Re-ID task via supervised fine-tuning, augmented with domain-agnostic prompts and a hard-sample mining strategy. Furthermore, a robust μ-distance metric is proposed to enable model-agnostic, plug-and-play re-ranking enhancement. Extensive experiments demonstrate consistent and significant performance gains across multiple DG Re-ID benchmarks, validating the method’s effectiveness and broad applicability.

Abstract

Domain Generalizable (DG) person re-identification (Re-ID) has attracted growing research interest due to its potential for deployment in unseen real-world scenarios. Most existing approaches address DG Re-ID by focusing on training domain-generalizable encoders but ignore the possible refinements in inference stage. In contrast, this work explores an alternative direction which improves inference re-ranking to enhance DG Re-ID. Conventional re-ranking methods typically rely on neighborhood-based distances to refine the initial ranking list, inherently depending on features produced by the Re-ID encoder. However, they deteriorate on target domains since the encoder lacks sufficient generalizability to produce reliable feature distances on unseen scenarios. Inspired by the remarkable generalization capabilities of recent Multimodal Large Language Models (MLLMs), we propose an MLLM-empowered distance metric to improve re-ranking in DG Re-ID. Specifically, we first adapt an MLLM to Re-ID data through supervised fine-tuning, which incorporates a domain-agnostic prompt and a query-candidate hard mining scheme. Then, the adapted MLLM is employed to compute a $μ$-distance during inference, which is robust to domain gap and significantly enhances subsequent re-ranking performance. Our approach is model-agnostic and can be seamlessly integrated into previous re-ranking frameworks. Extensive experiments demonstrate that our approach consistently yields substantial performance improvements across multiple DG Re-ID benchmarks. The code of this work will be released at https://github.com/RikoLi/MUSE soon.

Problem

• Domain Generalizable Re-ID
• re-ranking
• Multimodal Large Language Models
• inference refinement
• feature distance

Innovation

• Multimodal Large Language Models
• Domain Generalization
• Person Re-Identification
• Re-Ranking
• Inference-stage Enhancement