$\beta$-CLIP: Text-Conditioned Contrastive Learning for Multi-Granular Vision-Language Alignment

Fatimah Zohra, Chen Zhao, Hani Itani +1

CLIP achieves strong zero-shot image-text retrieval by aligning global vision and text representations, yet it falls behind on fine-grained tasks even when fine-tuned on long, deta…