Quantifying Interpretability in CLIP Models with Concept Consistency

Open in new window