Transporting the Cross-Modal Prototypes for Unsupervised Visible-Infrared Person Re-Identification.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Unsupervised Visible Infrared Person Reidentification (USVI-ReID) is challenging due to large cross-modality variance and lack of annotations.
• Generating reliable cross-modality labels and learning modality-invariant features are significant hurdles in USVI-ReID.

Purpose of the Study:

• To develop an effective unsupervised method for cross-modality person reidentification.
• To address the difficulties in generating labels and learning features for USVI-ReID.

Main Methods:

• Leveraging information from cross-modality inputs and predicted labels.
• Incorporating entropy minimization, uniform label distribution, and cross-modality matching.
• Designing a loop iterative training strategy with uniform prior guided optimal transport for prototype matching.

Main Results:

• The proposed method achieves high Rank-1 accuracy (69.4% on SYSU-MM01, 89.4% on RegDB) without annotations.
• The model demonstrates the ability to gradually self-learn useful information from unlabeled cross-modal data.
• Discriminative representations are generated for unlabeled cross-modal pedestrian data.

Conclusions:

• The developed approach effectively overcomes challenges in unsupervised visible infrared person reidentification.
• The method shows significant promise for practical applications requiring cross-modality pedestrian retrieval.
• The iterative training and matching strategy facilitates learning robust, modality-invariant features.