twinkle-ai/tw-drug-labels-vision
收藏资源简介:
tw-drug-labels-vision 是一份涵盖台湾食品药物管理署(TFDA)核发之44,663笔药品仿单/外盒的繁体中文多模态数据集。每一笔纪录同时包含PDF全部页面的渲染图(WebP多页)以及一份依统一17栏JSON Schema抽取自原始药品标示文件的结构化资料,可直接用于语言模型微调、视觉语言模型训练、文件问答、药品知识检索、繁体中文医药NLP任务之素材。
tw-drug-labels-vision is a Traditional Chinese multimodal dataset containing 44,663 drug package inserts and outer box labels issued by the Taiwan Food and Drug Administration (TFDA). Each record includes both multi-page WebP rendered images of all PDF pages and structured data extracted from the original drug labeling documents in accordance with a unified 17-field JSON Schema. It can be directly used as training resources for language model fine-tuning, visual-language model training, document question answering, pharmaceutical knowledge retrieval, and Traditional Chinese medical and pharmaceutical NLP tasks.




