Accuracy and Execution Time Trade-off Analysis of VGG 16, ResNet-50, and YOLOv8 CNN Models for Face and Eye Detection

Face and eye detection are essential tasks in computer vision with practical applications in security, biometric systems, and user authentication. Choosing the right model for these tasks can be tricky due to the balance needed between accuracy and speed. In this study, we took a closer look at three popular convolutional neural network models VGGI6, ResNet-50, and YOLOv8 across three different datasets: Labeled Faces in the Wild (LFW), CelebA, and the Face Detection Data Set and Benchmark (FDDB). Our findings show that CelebA achieved the highest accuracy across all models, followed by LFW, while FDDB performed the least effectively. YOLOv8 stands out as the best all-around choice, delivering top-notch accuracy (96.40% to 97.56<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>) with reasonable speed (0.41 to 0.46 seconds). VGG16 also achieves impressive accuracy (91.68% to 95.51 %) but takes a bit longer (0.70 to 0.76 seconds), making it ideal for situations where precision is key. On the other hand, ResNet-50 is the fastest (0.33 seconds), but its accuracy, especially for face detection, is lower, so it might be better suited for cases where speed is more important than perfect precision.