I have an image volume with shape (128,128,128) and mask volume with shape (128,128,128,1,6)
I want to train a network with spatial windows size (64,64,64), with uniform sampling.
If I enable a small data augmentation with scaling, I receive an assertion in the rand_spacial_scaling.py in the layer_op telling me the following:
"The interpolation orders should be specified for each input modality"
The assert is testing if the image.shape[-1] == len(interp_orders[field]), where field = the mask modality, image.shape is (128,128,128,1,6)
so it is testing if 6 == 1...
Adding more interp_orders in the mask modality doesn't work because it is an int field