MMVU
收藏资源简介:
MMVU数据集旨在评估专家级别的多学科视频理解能力。该数据集包含3000个专家标注的问答示例,涵盖了1529个专业领域的视频,涉及27个学科和四个主要领域(科学、医疗保健、人文与社会科学、工程)。每个示例都附有专家标注的推理理由和相关领域知识,旨在评估模型的答案正确性和推理质量。数据集的设计反映了现实世界专业领域的复杂性,确保了对领域知识的广度和推理深度的评估。
The MMVU dataset is designed to evaluate expert-level multidisciplinary video understanding capabilities. It contains 3,000 expert-annotated question-answering examples, covering 1,529 videos from professional domains that span 27 disciplines across four major fields: Science, Healthcare, Humanities and Social Sciences, and Engineering. Each example is paired with expert-annotated reasoning rationales and relevant domain-specific knowledge, aiming to assess both the correctness of model-generated answers and the quality of its reasoning processes. The dataset's design mirrors the complexity of real-world professional domains, ensuring evaluations of both the breadth of domain knowledge and the depth of reasoning abilities.




