1. An c++ implement of face detect, recognition and retrieve based on mtcnn, facenet and hnswlib.
2. Another web server implement based on Caffe and Flask can be found in Facenet-Caffe

1. OpenCV3.4.0
2. NCNN framework, use to detect face.
3. pretraned facenet model, can be downloaded in InceptionResnet.mlmodel

App builds and running on iPhone SE successfully, arm64 architecture.

1. Enroll and store features in local sandbox.
2. Retrieve by hnswlib.

3. All the name you enrolled can be duplicated, the retrieve algorithm identify the face when the Euclidean Distance is smaller than 0.5, since all the face embedding vector is under L2 norm and confined in a high dimensional sphere whose radius is euqal to 1. The largest distance is 2.0.