Identity-Compensated Style Distillation for Visible-Infrared Person Re-Identification.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Visible-Infrared Person Re-Identification (VI-ReID) faces challenges due to modality discrepancies and intra-class variations.
• Existing methods often sacrifice identity semantics during style alignment, neglecting fine-grained details.

Purpose of the Study:

• To propose an Identity-Compensated Style Distillation (ICSD) network for VI-ReID.
• To enhance discriminative power of modality-invariant features and improve cross-modality consistency.

Main Methods:

• The proposed ICSD network integrates a Style Knowledge Distillation (SKD) module and an Identity Discrimination Amplification (IDA) module.
• SKD uses Style Discrepancy Reduction (SDR) and Identity Knowledge Compensation (IKC) for style alignment and semantic preservation.
• IDA refines identity-specific cues to capture subtle inter-class differences.

Main Results:

• ICSD consistently outperforms state-of-the-art methods on SYSU-MM01, RegDB, and LLCM benchmarks.
• The study validates the effectiveness and complementarity of the SKD and IDA modules.
• The proposed method achieves improved accuracy in pedestrian matching across visible and infrared modalities.

Conclusions:

• The ICSD network offers a robust solution for Visible-Infrared Person Re-Identification.
• The approach effectively balances cross-modality style consistency with identity-specific feature discrimination.
• This work advances the field by addressing limitations in existing VI-ReID methodologies.