THU-CVML/ScienceQA_VLMEvalKit

ScienceQA是一个面向科学问题回答的多模态数据集，包含文本和图像两种模态。它旨在帮助模型学习通过不同模态的信息综合来进行科学问题的推理。数据集中的问题来源于小学和高中科学课程，涵盖了化学、生物学、物理学、地球科学、工程学、地理学、历史学等多个领域。每个问题都附带多个选择答案，以及正确答案的索引、提示、任务描述、年级水平、学科、主题、子类别、技能、相关讲座和解决方案等信息。数据集分为训练集、验证集和测试集，共计约2.7亿字节，包含超过2.1万个样本。

ScienceQA is a multimodal dataset for science question answering, which includes both text and image modalities. It is designed to help models learn to synthesize information from different modalities for scientific question reasoning. The questions in the dataset are sourced from elementary and high school science curricula, covering various fields such as chemistry, biology, physics, earth science, engineering, geography, history, and more. Each question comes with multiple choice answers, an index for the correct answer, hints, task descriptions, grade levels, subjects, topics, subcategories, skills, related lectures, and solutions. The dataset is split into training, validation, and test sets, totaling approximately 272 million bytes and containing over 21,000 samples.