m-a-p/SimpleVQA

SimpleVQA是一个多模态事实性评估数据集，用于评估多模态大型语言模型在回答自然语言简短问题时的事实性能力。该数据集包含多种任务和场景，确保了问题和答案的高质量与挑战性，同时具有静态且永不过时的参考答案，便于通过LLM-as-a-judge评分系统进行评估。

SimpleVQA is a multimodal factuality evaluation dataset for assessing the factuality ability of multimodal large language models in answering natural language short questions. It includes a variety of tasks and scenarios, ensuring high-quality and challenging questions, along with static and timeless reference answers, which are easy to evaluate through an LLM-as-a-judge scoring system.