Disparities In Negation Understanding Across Languages In Vision-Language Models

Create a library of negation-aware evaluation probes for VLMs across languages. This tool will help developers identify and debug the 'affirmation bias' that plagues current vision-language models when interpreting negative images or captions.

Suggested repo: not-a-vision

"Does your VLM struggle with 'not'?"

Estimated effort: 35h