# face_recognition_ios

**Repository Path**: samfky/face_recognition_ios

## Basic Information

- **Project Name**: face_recognition_ios
- **Description**: face recognition and retrieve in ios
- **Primary Language**: C++
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2019-08-06
- **Last Updated**: 2020-12-19

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# An ios app for face recognition and retrieve
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](https://github.com/taylorlu/Facenet-Caffe)
# Prerequisites
1. OpenCV3.4.0
2. [NCNN](https://github.com/Tencent/ncnn) framework, use to detect face.
3. pretraned facenet model, can be downloaded in [InceptionResnet.mlmodel](https://pan.baidu.com/s/1I7S-HHWwYcSxvMzK0RULwg)
# Test platform
  App builds and running on iPhone SE successfully, `arm64 architecture`.
# Application
1. Enroll and store features in local sandbox.
2. Retrieve by hnswlib.

<div align="center">
<img src="https://github.com/taylorlu/face_recognition_ios/blob/master/pics/IMG_1756.jpg" height="414" width="240" >
<img src="https://github.com/taylorlu/face_recognition_ios/blob/master/pics/IMG_1758.jpg" height="414" width="240" >
</div>

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.

<div align="center">
<img src="https://github.com/taylorlu/face_recognition_ios/blob/master/pics/centerloss.png" height="200" width="450" >
</div>