arxiv_cv 80% Match Research Paper Computer vision researchers,Security system developers,AI engineers,Surveillance analysts 2 weeks ago

FlexiReID: Adaptive Mixture of Expert for Multi-Modal Person Re-Identification

computer-vision › object-detection

📄 Abstract

Abstract: Multimodal person re-identification (Re-ID) aims to match pedestrian images across different modalities. However, most existing methods focus on limited cross-modal settings and fail to support arbitrary query-retrieval combinations, hindering practical deployment. We propose FlexiReID, a flexible framework that supports seven retrieval modes across four modalities: rgb, infrared, sketches, and text. FlexiReID introduces an adaptive mixture-of-experts (MoE) mechanism to dynamically integrate diverse modality features and a cross-modal query fusion module to enhance multimodal feature extraction. To facilitate comprehensive evaluation, we construct CIRS-PEDES, a unified dataset extending four popular Re-ID datasets to include all four modalities. Extensive experiments demonstrate that FlexiReID achieves state-of-the-art performance and offers strong generalization in complex scenarios.

Authors (8)

Zhen Sun

Lei Tan

Yunhang Shen

Chengmao Cai

Xing Sun

Pingyang Dai

+2 more

Submitted

October 17, 2025

arXiv Category

cs.CV

arXiv PDF

Key Contributions

FlexiReID is a flexible framework for multimodal person re-identification that supports arbitrary query-retrieval combinations across RGB, infrared, sketches, and text modalities. It utilizes an adaptive MoE mechanism and a cross-modal query fusion module, achieving state-of-the-art performance and strong generalization.

Business Value

Enhances security and surveillance capabilities by enabling accurate identification of individuals across different sensor types and even textual descriptions, improving operational efficiency.

Paper Metadata

Innovation Type

Algorithmic

Deployment Feasibility

Moderate. Requires robust feature extraction for multiple modalities and efficient MoE implementation.

Limitations Addressed

Limited cross-modal settings in existing methods,Inability to support arbitrary query-retrieval combinations,Hindrance to practical deployment

Performance Gains

State-of-the-art performance

Technical Tags

Multimodal person re-identificationcross-modal retrievaladaptive mixture-of-experts (MoE)feature extractionmodality fusionpedestrian matchingrgbinfraredsketchestext

Research Topics

Person re-identificationMultimodal learningComputer visionDeep learningFeature representation

Methods & Architectures

Adaptive Mixture-of-Experts (MoE)Cross-modal query fusionFeature integrationDeep learning models Mixture-of-Experts (MoE)

Applications & Tasks

Surveillance Security Retail analytics Smart cities Cross-modal matchingPedestrian identificationData heterogeneity Person re-identificationCross-modal retrievalPedestrian tracking

Datasets & Benchmarks

Datasets

CIRS-PEDES

Related Fields

Computer VisionMachine LearningPattern RecognitionSurveillance TechnologyMultimodal AI

Keywords

Person re-identificationMultimodalCross-modalMixture of ExpertsFeature extractionSurveillancePedestrianRGBInfraredSketchesTextRe-IDCIRS-PEDES

Academic Context

#Person re-identification#Multimodal learning#Computer vision#Deep learning#Feature representation

Commercial Potential

Potential Products

Multimodal surveillance systemsPerson tracking softwareSecurity analytics platforms

Target Industries

Security and SurveillanceRetailLaw EnforcementSmart CitiesTransportation

Use Case Examples

Tracking a suspect across different camera views (day/night)Identifying a person based on a sketch or descriptionEnhancing crowd monitoring systems

Competitive Edge

Offers greater flexibility and broader applicability than existing methods by supporting diverse modalities and arbitrary query-retrieval combinations.

Market Opportunity

Large and growing market for AI-powered security and surveillance solutions.

Revenue Models

Software licensingsystem integrationcloud-based analytics services.

Resource Requirements

Compute Needs

High, due to multimodal feature extraction and MoE complexity.

Data Requirements

Large-scale multimodal datasets with diverse pedestrian images across different sensors and modalities.

Deployment Constraints

Requires integration of multiple sensor types and robust feature extraction pipelines.

Scalability

Scalability depends on the efficiency of the MoE and feature fusion mechanisms.

Regulatory Considerations

Privacy concerns related to surveillance and person identification.

Production Readiness

Maturity Level

Research

Time to Market

2-4 years

Patent Potential

Moderate, for the adaptive MoE and cross-modal fusion techniques.

View Full Paper Back to Papers