Improving Object Detection Performance through YOLOv۸: AComprehensive Training and Evaluation Study

This study evaluated the performance of a YOLOv۸-based segmentation model fordetecting and segmenting wrinkles in facial images. The model's performance wasassessed using standard metrics, including Precision (P), Recall (R), and mean AveragePrecision (mAP) at thresholds of ۰.۵۰ (mAP۵۰) and ۰.۵۰–۰.۹۵ (mAP۵۰-۹۵), as well asMask Precision and Mask Recall to evaluate segmentation quality. The model was testedon a validation dataset of ۱۳۱ images, yielding a Precision of ۹۰.۷%, Recall of ۸۹.۱%,mAP۵۰ of ۸۷.۰%, and mAP۵۰-۹۵ of ۱۰.۲%. For segmentation, Mask Precision was ۸۰.۷%and Mask Recall was ۸۹.۱%. The model performed best in detecting forehead wrinkles,with a Precision of ۸۵.۰%, Recall of ۸۰.۷%, and mAP۵۰ of ۸۵.۷%. Detection of frownlines showed lower performance with a Precision of ۸۰.۵% and mAP۵۰ of ۸۱.۶%. Generalwrinkle detection achieved a Precision of ۸۸.۶%, but with a lower Recall (۸۱.۸%) andmAP۵۰ (۸۳.۷%). Although the model demonstrated strong localization and segmentationcapabilities, challenges were observed in detecting subtle wrinkles and handling complexlighting or overlapping features, resulting in false positives and under-segmentation insome cases.

کلیدواژه ها:

Object Detection ، YOLOv۸ ، Data Augmentation ، ModelOptimization ، Generalization