Fix installation, and some other bugs

[kushalkolar]

Hi,

This is a cool package! I had trouble getting it installed so I changed some things to make it easier for others.

I created a proper setup.py so that it can be easily installed via pip. For now it can be installed using pip install -e ., but you'll need to modify some things to get it working from any arbitrary dir if you want to have it on pypi.

About the installation:

1. scikit-image>0.15 do not have a min_size kwarg for skimage.morphology.remove_small_holes() which you use in utils.normalized.clean_image(). So the setup.py is set to work with 0.15.x.
2. Added __init__.py to model_layers and utils so that they're proper sub-packages that can be identified through setuptools.find_packages()
3. MANIFEST.in so that network model files etc. are included.

scikit-image 0.15 has the min_size kwarg: https://scikit-image.org/docs/0.15.x/api/skimage.morphology.html#skimage.morphology.remove_small_holes

it was removed in v0.16: https://scikit-image.org/docs/0.16.x/api/skimage.morphology.html?highlight=morphology%20remove_small_holes#remove-small-holes

Bug fixes:

1. Added usingCL attribute to NuSet class in NuSeT.py. Set it to False.
2. Set a bunch of variables to have default values of 0 in train_gui.py (line 189-196), otherwise line 272 will raise an exception. Not sure if this is the right way to fix this.

[kushalkolar]

Just to add onto this, there's a bug with tensorflow v1 w/regards to cuDNN and newer RTX cards. The following exception is raised when training or predicting:

2020-08-18 21:49:40.924336: E tensorflow/stream_executor/cuda/cuda_dnn.cc:329] Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR
2020-08-18 21:49:40.949069: E tensorflow/stream_executor/cuda/cuda_dnn.cc:329] Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR
  0%|                                                                                                                                                                                           | 0/3 [00:00<?, ?it/s]
Traceback (most recent call last):
  File "/share/python-virtual-environments/nuset/lib/python3.6/site-packages/tensorflow_core/python/client/session.py", line 1365, in _do_call
    return fn(*args)
  File "/share/python-virtual-environments/nuset/lib/python3.6/site-packages/tensorflow_core/python/client/session.py", line 1350, in _run_fn
    target_list, run_metadata)
  File "/share/python-virtual-environments/nuset/lib/python3.6/site-packages/tensorflow_core/python/client/session.py", line 1443, in _call_tf_sessionrun
    run_metadata)
tensorflow.python.framework.errors_impl.UnknownError: 2 root error(s) found.
  (0) Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
         [[{{node model_U-Net/conv1-1/Conv2D}}]]
         [[ToFloat/_105]]
  (1) Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
         [[{{node model_U-Net/conv1-1/Conv2D}}]]
0 successful operations.
0 derived errors ignored.

This seems to be a well known issue that can be fixed by just setting an environment variable: export TF_FORCE_GPU_ALLOW_GROWTH=true https://github.com/tensorflow/tensorflow/issues/24496