常见皮肤问题的图像精确识别和定位数据

数据包含了不同光照条件和拍摄角度下的图像，可用于医疗美容诊断、皮肤问题检测和个性化治疗方案制定相关的目标检测模型训练。通过精确的目标检测而非简单的图像分类，为医疗美容诊断和治疗方案制定提供更详细的皮肤问题线索。数据采集使用由AI模型生成的皮肤图像。收集涵盖不同肤色、年龄、性别、光照条件和拍摄角度的面容图像和临床面部图像。图像脱敏处理：眼睛区域屏蔽：将眼睛区域像素值置为0，图像裁剪：确保裁剪后的图像不包含眼睛。进行皮肤色值增强（皮肤明度调整和皮肤纹理）-随机旋转等图像增强处理。采用COCO格式进行多类目标检测标注，标注时用LabelImg工具，用包围框(x,y,width,height)精确定位每个皮肤问题。痤疮标注包括红肿部分；普通痣和异常痣主要靠颜色、形状和边缘规则性区分；粉刺标注包括周围轻微发红的区域；疤痕标注整个疤痕区域，包括周围略有变色的皮肤；色斑标注整个色素沉着区域。
将COCO格式转换为CSV格式：
x1=x,y1=y,x2=x+width,y2=y+height,cls∈{acne,mole1,mole2,pimple,scar,spot}分别为：痤疮、普通痣、异常痣、粉刺、疤痕、色斑。cls为类别，x1,y1,x2,y2是坐标。

This dataset comprises images captured under diverse lighting conditions and shooting angles, intended for training object detection models applied to medical aesthetic diagnosis, skin lesion detection and personalized treatment plan formulation. Unlike simple image classification, precise object detection can provide more detailed skin condition cues to support medical aesthetic diagnosis and treatment plan development.

Skin images generated by AI models are used for data collection. The collected samples include facial and clinical facial images covering various skin tones, age groups, genders, lighting conditions and shooting angles. Image de-identification processing is conducted as follows: first, eye region masking: set pixel values of eye regions to 0; second, image cropping: ensure cropped images do not contain eye areas. Various image augmentation operations are applied, including skin color enhancement (adjusting skin brightness and enhancing skin texture), random rotation and other related operations.

Multi-class object detection annotations are produced in COCO format using the LabelImg tool, with bounding boxes in the format of (x, y, width, height) to accurately locate each skin lesion. Specific annotation guidelines are listed below:
1. Acne annotations cover the red and swollen affected areas;
2. Common nevi and atypical nevi are distinguished primarily by color, shape and edge regularity;
3. Pimple annotations include the slightly reddened surrounding tissue;
4. Scar annotations cover the entire scar area, including the slightly discolored adjacent skin;
5. Spot annotations cover the entire hyperpigmented area.

Finally, the COCO-format annotations are converted to CSV format: x1 = x, y1 = y, x2 = x + width, y2 = y + height, where cls ∈ {acne, mole1, mole2, pimple, scar, spot}, corresponding to acne, common nevus, atypical nevus, pimple, scar and spot respectively. Here, cls denotes the category, and x1, y1, x2, y2 represent the bounding box coordinates.

搜集汇总

数据集介绍

特点

该数据集包含17076条皮肤问题图像数据，涵盖多种肤色、年龄和光照条件，适用于医疗美容诊断和皮肤问题检测。数据经过脱敏处理，采用COCO格式进行多类目标检测标注，支持精确的皮肤问题定位和识别。