# FOIL it! Find One mismatch between Image and Language caption

## TL;DR

This paper introduces FOIL-COCO, an extended dataset with subtle caption errors to evaluate if vision-language models truly understand image-text interactions, revealing current models' shortcomings compared to human performance.

## Contribution

The paper presents FOIL-COCO, a novel dataset with fine-grained caption errors, and demonstrates that current models struggle with tasks requiring detailed image-text understanding.

## Key findings

- Models perform poorly on caption classification, foil word detection, and correction.
- Humans achieve near-perfect accuracy on the same tasks.
- Current models rely insufficiently on visual cues for understanding image-caption relationships.

## Abstract

In this paper, we aim to understand whether current language and vision (LaVi) models truly grasp the interaction between the two modalities. To this end, we propose an extension of the MSCOCO dataset, FOIL-COCO, which associates images with both correct and "foil" captions, that is, descriptions of the image that are highly similar to the original ones, but contain one single mistake ("foil word"). We show that current LaVi models fall into the traps of this data and perform badly on three tasks: a) caption classification (correct vs. foil); b) foil word detection; c) foil word correction. Humans, in contrast, have near-perfect performance on those tasks. We demonstrate that merely utilising language cues is not enough to model FOIL-COCO and that it challenges the state-of-the-art by requiring a fine-grained understanding of the relation between text and image.