Acne Severity Detection and Classification: Comparing You Only Look Once Methods

Abstract

Acne (Acne vulgaris) is one of the most common skin diseases, especially on the face. Accurate diagnosis and proper treatment are important for optimal care results and improving the accuracy of detection and classification of acne severity. YOLO (You Only Look Once) is a deep learning method used for object detection in images. This study compares the results and performance of YOLOv5 and YOLOv8 in detecting acne on the face. Several experiments were also conducted with data pre-processing, model size, and the use of different basic hyperparameters on both models to understand the impact and differences between YOLOv5 and YOLOv8. The results show that YOLOv5 overall has higher performance in detecting acne compared to YOLOv8, which requires larger hyperparameter values and model sizes to achieve the most optimal results. Conservative hyperparameters (with relatively smaller values or sizes) on YOLOv5 contribute to better performance.