Dynamic Modality-Camera-Invariant Clustering for Unsupervised Visible-Infrared Person Re-Identification

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Unsupervised visible-infrared person re-identification (USL-VI-ReID) is a cost-effective alternative to supervised methods.
• Current USL-VI-ReID methods struggle with cross-camera variations, leading to identity splitting and reduced accuracy.

Purpose of the Study:

• To propose a novel dynamic modality-camera-invariant clustering (DMIC) framework for USL-VI-ReID.
• To eliminate cross-modality and cross-camera discrepancies in clustering for improved person re-identification.

Main Methods:

• The DMIC framework integrates modality-camera-invariant expansion (MIE), dynamic neighborhood clustering (DNC), and hybrid modality contrastive learning (HMCL).
• MIE fuses intermodal and intercamera distance coding.
• DNC refines optimization objectives for cross-modal and cross-camera generalizability, while HMCL optimizes instance- and cluster-level distributions.

Main Results:

• The proposed DMIC framework effectively addresses limitations in existing clustering approaches for USL-VI-ReID.
• DMIC achieves competitive performance, significantly reducing the gap with supervised methods.
• Experiments demonstrate improved accuracy and reliability in cross-modal associations.

Conclusions:

• The DMIC framework offers a robust solution for unsupervised visible-infrared person re-identification.
• This approach enhances the flexibility and cost-effectiveness of person re-identification systems.
• DMIC paves the way for more accurate and reliable person identification across different modalities and cameras.