When setting up the NiftyNet environment for use with a git-clone of the repo, using the recommended "pip install -r" method, one runs into the problem of the latest version of numpy (1.15) not being compatible with the version of TensorFlow that is to be installed (1.10). This leads to the install of TensorFlow failing.
The problem can be easily fixed by replacing the lines in requirements-{c,g}pu.txt that read: "numpy>=1.12" with "numpy==1.14.5" (the most recent version of numpy supported by TensorFlow 1.10).
When setting up the NiftyNet environment for use with a git-clone of the repo, using the recommended "pip install -r" method, one runs into the problem of the latest version of numpy (1.15) not being compatible with the version of TensorFlow that is to be installed (1.10). This leads to the install of TensorFlow failing. The problem can be easily fixed by replacing the lines in requirements-{c,g}pu.txt that read: "numpy>=1.12" with "numpy==1.14.5" (the most recent version of numpy supported by TensorFlow 1.10).