Real Time Face Recognition Formats

Real Time Face Recognition Formats, a powerful solution that combines the capabilities of MTCNN and MobileFacenet models using TensorFlow. This system is designed for seamless face detection and recognition, with support for both GPU and CPU processing in variety of formats. It offers features for converting models from Tensorflow native into TensorRT runtime and TensorFlow Lite format, enabling easy testing and deployment on edge devices.

System Requirements

• python>=3.9 (Recommend 3.9.13)
• opencv-python
• numpy==1.23.1 (fix booling type)
• scipy
• tf_slim
• scikit-learn
• scikit-image
• pycuda (Require cuda, python-dev in the OS)
• tensorflow([and-cuda] optional)

Running Guide

1. Save images of the individuals you want to recognize in face_db folder. Ensure that each image contains only one person and is named using the person's label, e.g., "Sunday.jpg."
2. cd ./face_recognition/
3. python camera_...demo.py

Performance

System specification: Xeon E3 1241v3, 16gb, GTX 1070

Run type	GPU	CPU	TRT	TFLite
FPS	50-60	40-50	50-60	30-40
Using GPU (%)	20-25	~	23-28	~
Using CPU (%)	15-20	35-40	20-25	10-15

Referencess

1. Face recognition model: MobileFacenet
2. MobileFacenet with Tensorflow
3. Face detection model: MTCNN
4. MTCNN with Tensorflow
5. TensorRT runtime
6. Tensorflow to TensorRT with Onnx
7. Tensorflow Lite