Install MXNET 1.5.X on Raspberry Pi 4

[Raunak-Singh-Inventor]

We must document the MXNET installation steps properly in Readme file.

The recommended way is to do this on any Linux machine with 16 GB memory and 8 CPU machine using Docker. It takes 15 mins to finish the build. Native building on raspberry itself is possible but takes more than 4-5 hours.

To install follow this link which has instructions but are not clear

https://mxnet.apache.org/versions/1.4.1/install/index.html?version=master&platform=Devices&language=Python&processor=CPU

Docker installation
Step 1 Install Docker on your machine by following the docker installation instructions.

Note - You can install Community Edition (CE)

Step 2 Post installation steps to manage [Docker] as a non-root user.

Follow the four steps in this docker documentation to allow managing docker containers without sudo.

Actual Build

Step 1: Go to folder of your choice e.g.

cd ~/Downloads

Step 2: Clone the repository e.g. targeting MXNET verison 1.5.x, using this command

git clone -b v1.5.x https://github.com/apache/incubator-mxnet.git mxnet

Step 3: go inside mxnet that you cloned

cd mxnet

Step 4: run this command to also get additional 3rd party repos for successful build

`git submodule update --init --recursive'

Step5: Run the build using docker by kicking of the python script

ci/build.py -p armv7

The resulting artifact will be located in build/mxnet-x.x.x-py2.py3-none-any.whl, copy this file to your Raspberry Pi. For us the target file name was mxnet-1.5.1-py2.py3-none-any.whl

You will also need to do this command on RaspberryPi

pip install mxnet-1.5.1-py2.py3-none-any.whl

sudo apt-get install libgfortran3

pip install --upgrade requests==2.20.1

pip install gluoncv==0.7.0

pip install opencv-python

[szha]

We have a guide on building for raspberry pi (source). This doc includes a guide on cross-compilation. Have you tried the instructions there? Do you see a need to update the doc?

[Raunak-Singh-Inventor]

@szha Yes! The instructions are very outdated and not beginner-friendly. I recommend updating it as per the above instructions. Also, it would make life easier if the compiled binary for various MxNet versions for raspberry pi and other 32-bit systems was provided at the PyPi project.