Real-Time ArcFace Multiplex Recognition

Face Detection and Recognition using RetinaFace and ArcFace, can reach nearly 24 fps at GTX1660ti.

How to run

• Install yarn
  • sudo apt install curl
  • curl -sS https://dl.yarnpkg.com/debian/pubkey.gpg | sudo apt-key add -
  • echo "deb https://dl.yarnpkg.com/debian/ stable main" | sudo tee /etc/apt/sources.list.d/yarn.list
  • sudo apt update && sudo apt install yarn
• Electron Node-JS Client
  • cd electron-client
  • yarn or npm install
  • yarn start or npm start
• Build R-CNN for Retina Face
  • cd ..
  • chmod a+x ./build_rcnn.sh
  • ./build_rcnn.sh
• Python Deal
  • python3 usb_camera.py -c X e.g: Replace X with 0
  • Click the corresponding Camera {X} Button at Electron

How to train mlp classifier

• mkdir ./Temp/raw
• mkdir ./Temp/train_data
• Place training pictures in the following format：

  ─── train_data
      ├── bush
      │   ├── 1559637960.1595788.jpg
      │   ├── 1559637960.1762984.jpg
      │   └── 1559637960.2001894.jpg
      ├── clinton
      │   ├── 1559637960.2104468.jpg
      │   ├── 1559637960.2225769.jpg
      │   └── 1559637960.281161.jpg
      └── obama
          ├── 1559637960.2940397.jpg
          ├── 1559637960.31212.jpg
          └── 1559637960.3381834.jpg

• python3 train_mlp.py

Verification

LResNet100E-IR network trained on MS1M-Arcface dataset with ArcFace loss:

Citation

@inproceedings{deng2020retinaface,
  title={Retinaface: Single-shot multi-level face localisation in the wild},
  author={Deng, Jiankang and Guo, Jia and Ververas, Evangelos and
          Kotsia, Irene and Zafeiriou, Stefanos},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={5203--5212},
  year={2020}
}

@inproceedings{deng2019arcface,
  title={Arcface: Additive angular margin loss for deep face recognition},
  author={Deng, Jiankang and Guo, Jia and Xue, Niannan and Zafeiriou, Stefanos},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={4690--4699},
  year={2019}
}