SciFIBench

SciFIBench是由剑桥大学创建的一个科学图表解释基准数据集，包含1000个多选题，分为两个任务和12个类别。数据集来源于arXiv计算机科学论文的图表和标题，通过对抗性过滤和人工验证确保难度和质量。SciFIBench旨在评估大型多模态模型在科学图表理解方面的能力，推动该领域的研究进展。

SciFIBench is a scientific chart interpretation benchmark dataset developed by the University of Cambridge. It comprises 1,000 multiple-choice questions, split into two tasks and 12 categories. The dataset is curated from charts and their accompanying titles in arXiv computer science papers, with adversarial filtering and manual validation employed to ensure its appropriate difficulty and quality. The core objective of SciFIBench is to evaluate the capabilities of large multimodal models in scientific chart understanding, and promote research progress in this field.