sunwayforever / mediapipe_demo

acomplish computer-vision tasks (object detection, face detection, face landmark, palm detection ... ) using mediapipe models

MediaPipe Demo

Getting started

• pip install numpy

• opencv

  • pip uninstall opencv-python

  • pip install opencv-python-headless

  • or build opencv from source in case you need cv.imshow

  opencv-python from pip was built with Qt support, which has version mismatch issue with pyqt5, we need to build opencv from source and remove Qt support

  https://forum.qt.io/topic/119109/using-pyqt5-with-opencv-python-cv2-causes-error-could-not-load-qt-platform-plugin-xcb-even-though-it-was-found

  https://docs.opencv.org/master/d7/d9f/tutorial\_linux\_install.html

  cmake -D PYTHON_DEFAULT_EXECUTABLE=$(python -c "import sys; print(sys.executable)") 
    -D PYTHON3_EXECUTABLE=$(python -c "import sys; print(sys.executable)") 
    -D PYTHON3_NUMPY_INCLUDE_DIRS=$(python -c "import numpy; print (numpy.get_include())") 
    -D PYTHON3_PACKAGES_PATH=$(python -c "from distutils.sysconfig import get_python_lib; print(get_python_lib())")
    -DOPENCV_EXTRA_MODULES_PATH=../opencv_contrib-master/modules ../opencv-master
  

• pip install pyqt5==5.15.2

• pip install zmq

• pip install scipy

• conda install cudnn==8.0.4 -c nvidia

• conda install cudatoolkit==11.1.74 -c nvidia

• pip install onnxruntime-gpu==1.7.0

• pip install onnx==1.8.1

• pip install tensorflow==2.4.0

or using a fresh installed conda environment:

• conda env create -f environment.yml

Run

$> cd python
$> ./launcher.sh

Note

flops:

• face detection: 52M
• face landmark: 56M
• iris landmark: 98M
• palm detection: 198M
• hand landmark: 710M
• face recognition: 2700M