MeMa

该数据集名为MeMa，总计包含42,307张图片，分为训练集（34,000张图片）、超参数选择集（3,729张图片）以及验证集（4,578张图片）。所有图片均被标注了疾病类别。MeMa数据集在避免侵犯真实患者隐私的同时，确保了丰富的且可信的医疗表现。经验丰富的临床专家对数据集进行了粗略和细致的标注。该数据集规模超过40,000张图片，其任务是针对医疗场景的面部去识别化。

The dataset is named MeMa, which contains a total of 42,307 images divided into three subsets: the training set (34,000 images), the hyperparameter selection set (3,729 images), and the validation set (4,578 images). All images are annotated with their corresponding disease categories. The MeMa dataset ensures rich and reliable medical manifestations while avoiding violations of real patients' privacy. Experienced clinical specialists conducted both coarse-grained and fine-grained annotations for this dataset. With over 40,000 images in total, the core task of the MeMa dataset is facial de-identification in medical scenarios.