Updating CLIP to Prefer Descriptions Over Captions

TL;DR

This paper enhances CLIP by fine-tuning it with the Concadia dataset to better differentiate between descriptive captions and full image descriptions, aligning more closely with blind and low-vision people's judgments.

Contribution

It introduces a parameter-efficient fine-tuning method for CLIP that emphasizes descriptions over captions, improving its suitability for accessibility and interpretability.

Findings

Better correlation with blind and low-vision judgments

Preserves transfer capabilities of CLIP

Provides interpretable model structure

Abstract

Although CLIPScore is a powerful generic metric that captures the similarity between a text and an image, it fails to distinguish between a caption that is meant to complement the information in an image and a description that is meant to replace an image entirely, e.g., for accessibility. We address this shortcoming by updating the CLIP model with the Concadia dataset to assign higher scores to descriptions than captions using parameter efficient fine-tuning and a loss objective derived from work on causal interpretability. This model correlates with the judgements of blind and low-vision people while preserving transfer capabilities and has interpretable structure that sheds light on the caption--description distinction.