Install PillowSIMD+libjpeg-turbo on Conda

gistfile1.txt

conda uninstall --force pillow -y

# install libjpeg-turbo to $HOME/turbojpeg
git clone https://github.com/libjpeg-turbo/libjpeg-turbo
pushd libjpeg-turbo
mkdir build
cd build
cmake .. -DCMAKE_INSTALL_PREFIX:PATH=$HOME/turbojpeg
make
make install
popd

# install pillow-simd with jpeg-turbo support
git clone https://github.com/uploadcare/pillow-simd
pushd pillow-simd
CPATH=$HOME/turbojpeg/include LIBRARY_PATH=$HOME/turbojpeg/lib CC="cc -mavx2" python setup.py install

# add turbojpeg to LD_LIBRARY_PATH
export LD_LIBRARY_PATH="$HOME/turbojpeg/lib:$LD_LIBRARY_PATH"

I couldn't do it without removing jpeg and libtiff packages.
Here is how I did it without messing with libpaths:

$ conda uninstall --force jpeg libtiff -y
$ conda install -c conda-forge libjpeg-turbo
$ CC="cc -mavx2" pip install --no-cache-dir -U --force-reinstall pillow-simd

@ivankreso, you probably want to reinstall libtiff after building pillow-simd, otherwise you may have some broken conda packages, since you removed their dependency. and while libjpeg-turbo replaces jpeg, the latter will still be re-installed by conda on the next update of any package relying on jpeg. So just as well, append to your suggestion:

conda install -y jpeg libtiff

Thanks to your input, I built a more complete version with quite a lot of prose explaining which is which here: https://docs.fast.ai/performance.html#faster-image-processing

Thanks for the gist soumith
I wouldnt uninstall any packages - just create a fresh conda, the minute that you start uninstalling stuff youll break everything.
a quick note. for libjpeg-turbo you need an asm compiler
sudo apt-get install nasm

I also had to install zlib:

sudo apt install zlib1g-dev