I created a new virtual environment in conda and I was successfully able to install caffe.

Its main core implementation is in C which got my attention when I started my Phd.

Other than C it also has wrappersinterface for Matlab, Python and command line.

Matlab interface is called matcaffe and python interface is called pycaffe.

In this post I will talk about my observations and experiences in installation process of pycaffe.

Caffe is one of the first frameworks I learned,modified,extended etc.

Actually main reason is that its C core and python interface makes it easy for me to use OpenCV with it which will be the next topic I will post.

This post is mainly about installation and the problems you can face during installation.

While it is not that hard to compile, you will need to install several dependencies to be able to successfully compile Caffe.

Also to compile pycaffe, you will need to have several packages installed.

You will get weird run time errors while you are even just trying to import Caffe in python.

Most probable causes to this problem are the libraries protobuf, readline,opencv.

If you will be compiling Caffe from source code, make sure there is only one version of these libraries in your system.

For example in linux, to check what version is available in runtime linker.

If you see that you have two versions of readline, things might get complicated.

Recent version of Caffe will need a recent version of protobuf library which might need a recent version of readline.

But old readline version in your system might be needed by many core parts of your distribution such as network-manager, ubuntu-desktop, wpa-supplicant etc.

Be careful if you are trying sudo apt-get remove libreadline for the old version, this might also remove some core parts in your system.

Check what will be automatically removed along with readline.

I have mistakenly removed my network-manager and wpasupplicant while trying to remove old libreadline version which caused me a lot of trouble to connect to the wireless networks to install them back again If you push more and do more crazy things, you might get a warning.

You can separately, in a new environment with Anaconda, install pycaffe.

To do that, you need to first create a new environment within Anaconda, this will create a bare minimum Anaconda environment without any additional libraries you installed.

After you activate the new environment, we will start installing libraries.

It should be able to install without any problem, it is a new clean environment.

After installation complete open a python console and check if everything is working.

Installing Caffe For Windows Python