Exploring Vision-Language Models for Imbalanced Learning