Mediapipe GPU for Python 3.8.0
Mediapipe docker images
HEY! 👋
I've made a docker container with mediapipe 0.10.7 and opencv 4.8.0 ready to use! Check it out.
If you don't want to use docker, keep reading.
Mediapipe 0.10.7 installation
Required time: 15min
Pre-requisites
Python 3.8.0
To install python 3.8.0 for l4t32.7.1 follow these steps, once done, come back here.
OpenCV
From the tests I've made, OpenCV is needed to make everything works correctly. Don't worry, I've already built it for you 😄
Follow the steps here to install opencv 4.8.0 from wheel. Then come back here.
Tensorflow
Wait, no tensorflow?
Well, I've seen guides that uses it and guides that don't. I didn't install it and my setup works great and uses the GPU.
The official mediapipe documentation says something about it here. I suggest you to continue without installing it.
Install mediapipe from wheel
First, update and upgrade:
sudo apt-get update && apt-get upgrade
Download the dependencies:
sudo apt-get install -y \
libopencv-core-dev \
libopencv-highgui-dev \
libopencv-calib3d-dev \
libopencv-features2d-dev \
libopencv-imgproc-dev \
libopencv-video-dev
Upgrade pip:
python3.8 -m pip install --upgrade pip
Download the wheel file from here.
Install mediapipe:
python3.8 -m pip install mediapipe-0.10.7-cp38-cp38-linux_aarch64.whl
To conclude, you need to install matplotlib:
python3.8 -m pip install matplotlib
To test that everything properly works, try these scripts.