Does CLIP Bind Concepts? Probing Compositionality in Large Image Models

Open in new window