Abstract: Recently, several generalist models such as Contrastive Language Image Pre-training (CLIP) have demonstrated their capabilities of performing diverse downstream tasks through zero-shot or ...