Visual USMLE
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/snap-stanford/med-flamingo
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含618个美国医学执照考试（USMLE）风格的题目，这些题目通过图像、病例摘要以及可能的实验室测量表格进行了增强，形成了一个具有挑战性的多模态问题集。此外，该数据集的特点是开放式问题而非选择题，这使得它在模型评估方面更具挑战性和现实性。规模上，数据集共有618个问题，任务类型为生成式视觉问答（Vqa）。
This dataset contains 618 questions modeled after the United States Medical Licensing Examination (USMLE). These questions are augmented with images, case summaries, and potentially laboratory measurement tables, forming a challenging multimodal question set. Furthermore, the dataset features open-ended questions rather than multiple-choice ones, which makes it more challenging and realistic for model evaluation. In terms of scale, the dataset has a total of 618 questions, and the task type is generative Visual Question Answering (VQA).
提供机构：
Adapted from the Amboss platform
