Installing NumPy and OpenCV from source

If for some reason you decided not to use Anaconda and are on an non-x86 Architecture where no prebuild packages for NumPy and OpenCV are available, than you can roughly follow this to build these packages from source.

Sidenote: I did this because at this point I hadn’t opted for anaconda yet. But I did the switch later, making the steps kinda pointless. Don’t be like me, make an educated decision before starting :)

NumPy

Again there is no reason to do this on standard architectures with internet access available

You first need to have Cython installed to need to install it from source. Download the Cython sources from PyPi and extract them. In the sources-folder run this (assuming you have at least a standard C compiler installed)

python setup.py install

(please make sure python actually points to a Python 3 excutable this might not always be the case)

This might require the permissions to write to some system-directories. If this is the case there is a way around it, but bash-history is available anymore.

Now you need the NumPy sources

git clone https://github.com/numpy/numpy

And build NumPy inside the sources folder

python setup.py build_ext --inplace

You might need to specify where Cython is installed to. But these are things lost in the shell :(

OpenCV

OpenCV requires NumPy so you need to have it installed first.

git clone https://github.com/opencv/opencv.git
mkdir build
cd build

And now simply run

cmake ../
make
make install

By default this will try install system-wide. You can change this behavior but again, thing lost in the shell.