MEDIQA-M3G

该数据集名为MEDIQA-M3G，包含了皮肤科领域的多语言和多模态医疗答案生成样本，每个样本包括医疗图像、文本查询以及多个可能的回应及其对应分数。该数据集提供中文、英文和西班牙语版本，非英文的训练集由机器翻译，而验证集和测试集则由人工翻译以确保准确性。每个实例采用JSON格式，包含唯一的遭遇ID和一系列图像ID。在规模上，训练集包含842个实例，验证集有56个实例，测试集有100个实例。该数据集的任务是医疗问题回答。

This dataset, named MEDIQA-M3G, consists of multilingual and multimodal medical answer generation samples in the field of dermatology. Each sample includes medical images, text queries, multiple candidate responses and their corresponding scores. The dataset is provided in Chinese, English and Spanish versions: the non-English training splits are machine-translated, while the validation and test sets are human-translated to ensure accuracy. Each instance is formatted in JSON, containing a unique encounter ID and a list of image IDs. In terms of scale, the training set contains 842 instances, the validation set has 56 instances, and the test set includes 100 instances. The task supported by this dataset is medical question answering.