TensorFlow is an amazing tool for neural networks. The only thing we don’t like is the need for bazel. The files get huge, you have to work in the TensorFlow repo and it is just annoying. Also the integration with other projects is really difficult. So we searched for a solution and we found one. Actually we didn’t, but we found an amazing GitHub repo. You can find it here.
In my case I had to checkout Version 0.12 for the installing of Protobuf and Eigen, but compiled the latest TensorFlow Version as a library. Everything works perfect. Connor Weeks also provides a few examples and the CMakeLists.txt file. For basic usage this TensorFlow library is enough, but if you need other functions, e.g. to serve a model, just modify the build rule and add the needed dependencies.