vlm-modality-research/chartqa-evidence-conflict-v1

ChartQA Evidence Conflict This dataset contains 230 reviewed conflicts between a native ChartQA chart and an evidence-bearing textual report. The original chart supports one answer, while counterfactual facts in the report support a distinct answer to the same question. Neither source is privileged in the evaluation prompt. Each row contains the original chart, shared question, chart-supported answer, report-supported answer, evidence-bearing report, unit and counterfactual… See the full description on the dataset page: https://huggingface.co/datasets/vlm-modality-research/chartqa-evidence-conflict-v1.