DermaBench
收藏资源简介:
DermaBench是一个由临床医生标注的皮肤病视觉问答基准数据集,基于多样皮肤病图像(DDI)数据集构建,包含656张来自570名不同患者的临床图像,覆盖Fitzpatrick皮肤类型I-VI。数据集通过22个主要问题(单选、多选和开放式)进行标注,涵盖诊断、解剖部位、病变形态等多个维度,共产生约14,474条VQA式标注。数据集的创建过程包括多轮专家评审和共识制定,确保了标注的临床准确性和一致性。DermaBench旨在评估视觉语言模型在皮肤病学中的视觉理解和推理能力,为多模态AI研究提供可靠基准。
DermaBench is a clinical-annotated dermatology visual question answering (VQA) benchmark dataset constructed based on the Diverse Dermatological Image (DDI) dataset. It contains 656 clinical images from 570 unique patients, covering Fitzpatrick skin types I to VI. The dataset is annotated with 22 main question types including single-choice, multiple-choice and open-ended questions, spanning multiple dimensions such as diagnosis, anatomical location, lesion morphology and more, totaling approximately 14,474 VQA-style annotations. The dataset creation process involves multiple rounds of expert review and consensus formulation to guarantee the clinical accuracy and consistency of the annotations. DermaBench is designed to evaluate the visual understanding and reasoning abilities of vision-language models in the field of dermatology, providing a reliable benchmark for multimodal AI research.




