Visible-Infrared Person Re-Identification via Partially Interactive Collaboration

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Visible-infrared person re-identification (VI-ReID) faces challenges due to significant cross-modality discrepancies between visible and infrared images.
• Existing two-stream network methods often overlook the crucial interaction between feature extraction processes of different modalities.

Purpose of the Study:

• To propose a partially interactive collaboration (PIC) method to reduce the modality gap in VI-ReID.
• To leverage complementary information from different modalities for enhanced person re-identification.

Main Methods:

• A partially interactive-shared architecture with collaborative shallow layers and shared deep layers is introduced.
• Collaborative shallow layers facilitate mutual constraint between modality-specific features, enhancing representations.
• Shared deep layers embed features into a common space for identity discriminability, trained with conventional and collaborative losses.

Main Results:

• The proposed PIC method demonstrates superior performance on public VI-ReID datasets.
• Achieved rank-1 accuracy of 83.6% on RegDB and 57.5% on SYSU-MM01 datasets.

Conclusions:

• The PIC method effectively exploits cross-modality information through interactive learning.
• This approach significantly improves the performance of visible-infrared person re-identification systems.