Face Recognition Demo consists of two models, RetinaFace and FaceNet. It can detect face on image and recognize who it is. Here are two judgment indicators, cosine similarity and Euclidean distance. The closer the cosine similarity is to 1 and the closer the Euclidean distance is to 0, the more similar is between two faces.

Here takes lin_1.jpg as example. Inference results on Edge2. Only show cosine similarity.

Inference speed test: USB camera about 44ms per frame. MIPI camera about 33ms per frame.

Clone the source code from our khadas/edge2-npu.

$ git clone https://github.com/khadas/edge2-npu

$ sudo apt update
$ sudo apt install cmake libopencv-dev

Like facenet, there are also two modes of this demo. One is converting face images into feature vectors and saving vectors in the face library. Another is comparing input face image with faces in the library and outputting Euclidean distance and cosine similarity.

Put retinaface.rknn and facenet.rknn in edge2-npu/C++/retinaface/data/model.

# Compile
$ bash build.sh
 
# Run mode 1
$ cd install/face_recognition
$ ./face_recognition data/model/retinaface.rknn data/model/facenet.rknn 1

After running mode 1, a file named face_feature_lib will generate in edge2-npu/C++/face_recognition/install/face_recognition/data. With this file generated, you can run mode 2.

# Run mode 2
$ ./face_recognition data/model/retinaface.rknn data/model/facenet.rknn data/img/lin_1.jpg

Put retinaface.rknn and facenet.rknn in edge2-npu/C++/face_recognition_cap/data/model.

# Compile
$ bash build.sh

Put your full face photo into the edge2-npu/C++/face_recognition/data/img. Recompile and run mode 1 to generate face_feature_lib. Put face_feature_lib in edge2-npu/C++/face_recognition_cap/install/face_recognition_cap/data.

# Run USB camera
$ cd install/retinaface_cap
$ ./face_recognition data/model/retinaface.rknn data/model/facenet.rknn usb 60
 
# Run MIPI camera
$ cd install/retinaface_cap
$ ./face_recognition data/model/retinaface.rknn data/model/facenet.rknn mipi 42