JetsonNano-ArcFace

A high-performance face recognition application with GUI on Jetson Nano (6-9fps)/PC (>15ps)

I. Required:

• LibTORCH for C++ (PC/Jetson)
• OpenCV 3.4.x with CUDA support (Not working with opencv 4.x.x)
• Cuda/CuDNN/TensorRT (PC)
• Qt5-GUI:

  sudo apt -y install cmake cmake-gui cmake-qt-gui
  sudo apt -y install qtbase5-dev
  sudo apt-get -y install qtdeclarative5-dev
  
  sudo apt -y install libcrypto++-dev 
  sudo apt -y install libgoogle-glog-dev 
  sudo apt -y install libgflags-dev
  sudo apt -y install --no-install-recommends libboost-all-dev

If you are using Jetson you don't have to install Cuda/CuDNN/TensorRT.

II. Clone & Run

Step 1: Clone project:

git clone https://github.com/CuteBoiz/JetsonNano-ArcFace
cd JetsonNano-ArcFace

Step 2: Edit CmakeLists.txt

If you are using Jetson Nano: rm CMakeLists.txt && mv CMakeLists.txt.jetson-nano CMakeLists.txt

gedit CMakeLists.txt 

• Change my username (tanphatnguyen) to your username
• Change libtorch directory (line 10) (PC only)
• Change TensorRT version (line 17 & 18) (PC only)

Step 3: Compile & Run:

cmake .
make
./main

References

• MTCNN
• Arcface
• MobileFaceNets
• libSVM

To-Do

• Convert all engine to onnx.
• Add export TensorRT engine.
• Add Non GUI Inference code.