MicroVQA
收藏资源简介:
MicroVQA是一个针对生物显微镜学领域视觉问答的基准数据集,由斯坦福大学等多个研究机构的生物专家手工创建了1042个视觉问答样本。数据集中的问题涵盖了专家图像理解、假设生成和实验提议三个关键的科学探究任务,旨在评估多模态大型语言模型在科学研究中的推理能力。数据集的问题和答案都是由专家编写的,涉及从细胞形态到技术成像挑战等多个方面,使用了多种显微镜模态,覆盖了从组织到原子级别的不同尺度,并以人类和鼠标等与人类生物学和医学相关的研究为主。
MicroVQA is a benchmark dataset for visual question answering (VQA) in the field of biological microscopy. It contains 1042 manually curated visual question answering samples created by biological experts from multiple research institutions including Stanford University. The questions in the dataset cover three core scientific inquiry tasks: expert image understanding, hypothesis generation, and experimental proposal, aiming to evaluate the reasoning capabilities of multimodal large language models (LLMs) in scientific research. Both the questions and answers in the dataset are expert-written, covering various aspects from cell morphology to technical imaging challenges. It utilizes multiple microscopy modalities, spans scales ranging from tissue to atomic-level, and primarily focuses on research related to human biology and medicine, such as studies on humans and mice.




