DW-ReID: Vision-Language Learning for Person Re-Identification Under Diverse Weather Conditions.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Person re-identification (ReID) performance significantly degrades in adverse weather conditions like rain, snow, or haze.
• Existing ReID methods and conventional image restoration techniques are often ill-suited for the high-level task of identity discrimination in challenging environments.

Purpose of the Study:

• To develop a unified framework, DW-ReID, that integrates weather-degraded image restoration with person re-identification.
• To improve the robustness and accuracy of person ReID systems under diverse and adverse weather conditions.

Main Methods:

• A two-stage training paradigm utilizing a large-scale Contrastive Language-Image Pre-training (CLIP) model.
• Optimization of learnable text prompts to create identity-specific descriptions.
• End-to-end joint training of a weather encoder, image restorer, and ReID encoder using language supervision.

Main Results:

• Demonstrated effectiveness and superior performance of the DW-ReID method on newly contributed synthetic datasets.
• Successful integration of image restoration and ReID tasks within a single framework.
• Validation of the language supervision approach for training weather-robust ReID models.

Conclusions:

• The proposed DW-ReID framework effectively addresses the challenge of person re-identification in diverse weather conditions.
• The integration of weather restoration and ReID through language supervision offers a promising direction for robust visual recognition systems.
• DW-ReID achieves state-of-the-art performance, outperforming existing methods in adverse weather scenarios.