MM-Skin

该数据集是首个大规模的多模态皮肤病学数据集，涵盖了三种成像方式（临床、皮肤镜和病理），包含了近1万份从专业教材中收集的高质量图像与文本配对。此外，数据集还包含了额外生成的遵循指令的视觉问答（VQA）样本（2.7万份），并在与其他皮肤病学数据集的比较中进行了基准测试。该数据集规模宏大，拥有1万份图像与文本配对，以及43,000张用于多数据集评估的总图像。其任务涵盖了视觉问答（VQA）和针对分类任务的监督微调（SFT）。

This dataset is the first large-scale multimodal dermatology dataset covering three imaging modalities: clinical, dermoscopic, and pathological. It contains nearly 10,000 high-quality image-text pairs collected from professional textbooks. Additionally, the dataset includes 27,000 additional instruction-following visual question answering (VQA) samples, and has been benchmarked against other dermatology datasets. With a substantial scale, this dataset features 10,000 image-text pairs and a total of 43,000 images for multi-dataset evaluation. Its supported tasks include visual question answering (VQA) and supervised fine-tuning (SFT) for classification tasks.