Test / Add GPU support for ClasssifyPixels-Unet module

[cells2numbers]

The module ClasssifyPixels-Unet uses keras with CNTK (windows) or tensorflow (Mac/Linux) backend with CPU support.

It takes 10 seconds on a MacBook Pro and ~70seconds on a Windows laptop but has not been tested with GPU support.

• GPU support for Mac/Linux should be simple and could be integrated using

configuration = tf.ConfigProto()
configuration.gpu_options.allow_growth = True
configuration.gpu_options.visible_device_list = "2"
session = tf.Session(config = configuration)

• For CNTK we would have to install a new package pip install cntk-gpu . This could be realized using a third requirements-windows-gpu.txt that replaces the current requirements-windows.txt.