Unable to run train_synthesis.py

[Salamander321]

In provider.py at line 111

data.transform.MinMaxNormalization(tf.uint16.max), There should be this

data.transform.MinMaxNormalization(),

In paper you have mentioned that you are able to perform one iteration of auto-context model so i was trying to first run one iteration first. But while implementing i got following errors

tensorflow.python.framework.errors_impl.InvalidArgumentError: 2 root error(s) found.
  (0) Invalid argument: Paddings must be non-negative: -86 -86
	 [[{{node compose/center_pad_3d/Pad}}]]
	 [[patch/IteratorGetNext]]
	 [[patch/IteratorGetNext/_1755]]
  (1) Invalid argument: Paddings must be non-negative: -86 -86
	 [[{{node compose/center_pad_3d/Pad}}]]
	 [[patch/IteratorGetNext]]
0 successful operations.
0 derived errors ignored.

shape passed to center_pad_3d transformation function is Volume_shape : [260, 340, 360, 1]

Is preprocessing mechanism for contex-aware is different than that of unet,pixtopix one ?

[bodokaiser]

data.transform.MinMaxNormalization(tf.uint16.max), There should be this data.transform.MinMaxNormalization(),

Why?

Is preprocessing mechanism for contex-aware is different than that of unet,pixtopix one ?

Yes, it is very different. For the context-aware network you need 3d patches while the other ones use 2d slices. I was never able to achieve good results with the context-aware network. The original paper is (in my opinion) not very specific on the implementation details and you have to write a lot of custom code for your training procedure.