Updating CLIP to Prefer Descriptions Over Captions
This work addresses a specific shortcoming in image-text similarity metrics for accessibility applications, representing an incremental improvement.
The authors tackled the problem that CLIPScore fails to distinguish between captions and descriptions, which is crucial for accessibility, by fine-tuning CLIP on the Concadia dataset to prefer descriptions, resulting in a model that correlates with blind and low-vision people's judgments while maintaining transfer capabilities.
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.