Explaining CLIP Zero-shot Predictions Through Concepts

TL;DR

EZPC explains CLIP's zero-shot image recognition by projecting its embeddings into a human-understandable concept space, maintaining accuracy and enhancing interpretability without extra supervision.

Contribution

Introduces EZPC, a method that bridges CLIP's predictions with interpretable concepts using a learned projection, without requiring additional concept labels.

Findings

Maintains CLIP's zero-shot accuracy on benchmark datasets.

Provides meaningful concept-level explanations for predictions.

Grounds open-vocabulary predictions in explicit semantic concepts.

Abstract

Large-scale vision-language models such as CLIP have achieved remarkable success in zero-shot image recognition, yet their predictions remain largely opaque to human understanding. In contrast, Concept Bottleneck Models provide interpretable intermediate representations by reasoning through human-defined concepts, but they rely on concept supervision and lack the ability to generalize to unseen classes. We introduce EZPC that bridges these two paradigms by explaining CLIP's zero-shot predictions through human-understandable concepts. Our method projects CLIP's joint image-text embeddings into a concept space learned from language descriptions, enabling faithful and transparent explanations without additional supervision. The model learns this projection via a combination of alignment and reconstruction objectives, ensuring that concept activations preserve CLIP's semantic structure while remaining interpretable. Extensive experiments on five benchmark datasets, CIFAR-100, CUB-200-2011, Places365, ImageNet-100, and ImageNet-1k, demonstrate that our approach maintains CLIP's strong zero-shot classification accuracy while providing meaningful concept-level explanations. By grounding open-vocabulary predictions in explicit semantic concepts, our method offers a principled step toward interpretable and trustworthy vision-language models. Code is available at https://github.com/oonat/ezpc.