VeriSciQA

VeriSciQA is a large-scale, high-quality dataset for Scientific Visual Question Answering (SVQA), containing 20,351 QA pairs spanning 20 scientific domains and 12 figure types. The dataset is constructed using a Generate-then-Verify framework that leverages cross-modal consistency checks to promote correctness of generated question-answer pairs.

VeriSciQA是一款面向科学视觉问答（Scientific Visual Question Answering，SVQA）的大规模高质量数据集，包含20351组问答对，覆盖20个科学领域与12种图表类型。该数据集采用「生成-验证」（Generate-then-Verify）框架构建，借助跨模态一致性校验机制保障生成问答对的正确性。