MEDHEB/MedHEB-Bench

该存储库托管了完整的MedHEB-Bench基准测试——一个多模态医学嵌入/检索基准，涵盖2D成像、3D成像和生物医学文本。它包括用于测试和训练任务的JSON和JSONL格式数据记录，涉及多种医学图像数据集（如胸部X光、皮肤疾病图像、脑MRI等）和文本数据集（如医学问答）。数据集结构分为2D任务、3D任务和文本任务，并包含预处理脚本。部分图像档案由于上游许可限制未重新分发，但任务定义JSON文件保留以展示基准结构。This repository hosts the complete MedHEB-Bench benchmark, a multimodal medical embedding/retrieval benchmark covering 2D imaging, 3D imaging, and biomedical text. It includes JSON and JSONL formatted data records for test and training tasks, spanning multiple medical image datasets such as chest X-rays, skin disease images, brain MRIs, etc., and text datasets such as medical question answering. The dataset structure is divided into 2D tasks, 3D tasks, and text tasks, and also includes preprocessing scripts. Some image archives are not redistributed due to upstream licensing restrictions, but the task definition JSON files are retained to demonstrate the benchmark's structure.