DermaSynth
收藏资源简介:
DermaSynth数据集是一个包含丰富的合成图像-文本对的皮肤科数据集。它利用公开的皮肤科数据集创建了图像和问题的配对,以及相应的答案。这个数据集特别适用于皮肤病变的问题回答任务,包含了训练集的划分信息,共有约9.2万个示例。
The DermaSynth dataset is a dermatology dataset that contains a rich collection of synthetic image-text pairs. It constructs paired images, questions and their corresponding answers based on publicly available dermatology datasets. This dataset is specifically tailored for skin lesion question answering tasks, and includes training set split information, with a total of approximately 92,000 examples.
数据集概述
数据集名称
DermaSynth
语言
- 英文 (en)
许可
cc-by-nc-4.0
数据规模
- 10K < n < 100K
任务类别
- 问题回答 (question-answering)
数据特征
- image_id: 字符串 (string)
- image_path: 字符串 (string)
- dataset_origin: 字符串 (string)
- question: 字符串 (string)
- answer: 字符串 (string)
数据拆分
- 训练集 (train)
- 文件大小: 43420729 字节
- 示例数量: 92020
下载大小
14384301 字节
数据集大小
43420729 字节
配置
- 默认配置 (default)
- 数据文件
- 拆分: 训练集 (train)
- 路径: data/train-*
- 数据文件
标签
- 皮肤科 (dermatology)
- 皮肤病变 (skin_lesion)
相关链接
- GitHub 链接: GitHub - abdurrahimyilmaz/DermaSynth
- 论文: DermaSynth: Rich Synthetic Image-Text Pairs Using Open Access Dermatology Datasets
- 微调模型: Hugging Face - abdurrahimyilmaz/DermatoLlama-1.0
引用信息
bibtex @article{yilmaz2025dermasynth, title={DermaSynth: Rich Synthetic Image-Text Pairs Using Open Access Dermatology Datasets}, author={Yilmaz, Abdurrahim and Yuceyalcin, Furkan and Gokyayla, Ece and Choi, Donghee and Erdem, Ozan and Demircali, Ali Anil and Varol, Rahmetullah and Kirabali, Ufuk Gorkem and Gencoglan, Gulsum and Posma, Joram M and Temelkuran, Burak}, journal={arXiv preprint arXiv:2502.00196}, year={2025} }




