CLIP Behaves like a Bag-of-Words Model Cross-modally but not Uni-modally

VLM
Re-assign community