Skin Lesion and Healthy Skin Dataset for 4-Class Instance Segmentation (YOLO annotation)

This dataset was built to carry out the research described in the paper titled "Segmentation of Skin Lesions Using Deep YOLO-Family Networks: A Comparison of the Performance of Selected Models on a New Dataset". The full dataset consists of 7,000 images (after augmentation) curated from the ISIC archive. It is composed of 4 distinct skin lesion classes (2 benign: Melanocytic nevus (N) and Pigmented benign keratosis (PBK); 2 malignant: Invasive melanoma (MI) and a combined class of Basal cell carcinoma and Squamous cell carcinoma (BCC+SCC) ) and a dedicated 'background' (BGD) class containing images of clear, healthy skin without any lesions. The dataset is balanced through the augmentation of the MI, N, and PBK classes belonging solely to the training set. The dataset was randomly divided into training, validation, and test sets, with the validation and test sets containing exactly 100 images for each class including background images. Detailed annotations of the skin lesion areas were created in the form of polygons in the format required by YOLO-family models. The dataset is particularly useful for experiments in automated skin lesion segmentation and can be used to train and benchmark instance segmentation models (like YOLOv8, YOLOv9, YOLOv11, YOLOv12, and YOLOv26) , with a specific focus on enhancing robustness and reducing false-positive detections on healthy skin.

Using this dataset please cite:

Omiotek, Z., Krukar, N., Olejarz, A., Lichograj, P., Komada, M., & Konieczna, M. (2026). Segmentation of Skin Lesions Using Deep YOLO-Family Networks: A Comparison of the Performance of Selected Models on a New Dataset. Electronics, 15(8), 1545. https://doi.org/10.3390/electronics15081545