metadata
license: cc-by-4.0
Introduction
TypoEval Dataset is a large-scale dataset for evaluating typographic threats against Vision-Language Foundation Models. It contains a large collection of typographic examples organised into two sub-datasets (TypoWD and TypoID) and their invisible counterparts (TypoWD-Invisible and TypoID-Invisible).
🚧 Status. This is a pre-release version accompanying a manuscript currently under review. Field names and directory layout may change. Author information, the paper link, and the evaluation code will be added once the manuscript is accepted.