Evaluating Vision-Language Models on Misleading Data Visualizations (Dataset)
Overview
This dataset accompanies the paper:
“When Visuals Aren’t the Problem: Evaluating Vision-Language Models on Misleading Data Visualizations.”
MisVisBench is designed to evaluate whether Vision-Language Models (VLMs) can detect misleading information in data visualization-caption pairs, and whether they can correctly attribute the source of misleadingness to appropriate error… See the full description on the dataset page: https://huggingface.co/datasets/MaybeMessi/MisVisBench.
