Comparison of ArcFace and Dlib Performance in Face Recognition with Detection Using YOLOv8

Abstract

This study compares the performance of ArcFace and Dlib models in face recognition with YOLOv8 used for face detection on a limited dataset. The evaluation used metrics such as accuracy, F1-score, recall, and precision. ArcFace, which employs the Additive Angular Margin Loss method, demonstrated superior performance with the highest accuracy of 0.90, precision of 0.90, recall of 1.00, and an F1-score of 0.95. Meanwhile, Dlib achieved an accuracy of 0.57, precision of 0.57, recall of 1.00, and an F1-score of 0.73. The aim of the study was to find the best model in terms of accuracy. ArcFace proved to be more accurate and suitable for applications requiring high reliability, such as advanced security systems, identity verification, and research that demands high precision in face recognition. Dlib, although less accurate, offers speed and simplicity, making it suitable for rapid prototyping and lightweight applications with limited resources. The results indicate that ArcFace outperforms in face recognition on limited datasets, while Dlib is more appropriate for simple applications requiring lightweight computation. This study provides guidance for developers in selecting the appropriate face recognition model to meet specific needs in both industry and research.