SmartCLIP: Modular Vision-language Alignment with Identification Guarantees

arXiv:2507.22264v2 Announce Type: replace-cross Abstract: Contrastive Language-Image Pre-training (CLIP)~\citep{radford2021learning} has emerged as a pivotal model in computer vision and multimodal learning, achieving state-of-the-art performance at aligning visual and textual representations throug