PERFORMANCE COMPARISON OF FACENET PYTORCH AND KERAS FACENET METHODS FOR MULTI FACE RECOGNITION

Face recognition has become an important technology in various applications, but challenges arise when multiple faces must be recognized simultaneously in a single image or video frame. This study develops a multi-face recognition system using the Multi-Task Cascaded Convolutional Neural Network (MTCNN) method for face detection, Pytorch Facenet and Keras Facenet for recognition, and Support Vector Machine (SVM) for classification. Using a dataset of 1000 images from 10 classes, this study compares the performance of Pytorch Facenet and Keras Facenet in terms of speed, memory usage, and accuracy. The results show that Pytorch Facenet is faster with an average of 0.15 seconds per image compared to Keras Facenet which requires 0.86 seconds per image, and is more efficient in memory usage with 384.19 MB lower. However, Pytorch Facenet uses 3% more CPU. In addition, in terms of accuracy, Pytorch Facenet shows a more stable and consistent confidence score. In conclusion, Pytorch Facenet proves to be more efficient and reliable for multi-face recognition, although it requires further CPU optimization for more optimal use in real application scenarios.