OR-VSKC

该数据集包含超过34,000张由扩散模型生成的合成图像，这些图像描绘了手术室场景，其中包含违反既定安全规则的实体。此外，数据集还包含了214张由人类标注的图像，这些图像作为验证的黄金标准。该数据集旨在揭示并研究多模态大型语言模型中的视觉语义知识冲突，特别是在手术安全背景下。数据集的规模为34,000张合成图像和214张人类标注图像，其任务是对手术风险进行检测，并评估视觉语义知识冲突。

This dataset contains over 34,000 synthetic images generated by diffusion models, which depict operating room scenes featuring entities that violate established safety protocols. Additionally, the dataset includes 214 human-annotated images serving as the gold standard for model validation. This dataset aims to uncover and investigate visual semantic knowledge conflicts in multimodal large language models, specifically within the context of surgical safety. Comprising 34,000 synthetic images and 214 human-annotated images, the dataset is designed for the tasks of surgical risk detection and visual semantic knowledge conflict evaluation.