Which datatype to use saving nifti labels

[davidkvcs]

Hello

In your guide (https://github.com/microsoft/InnerEye-DeepLearning/blob/main/docs/creating_dataset.md) you write:

• For images can be encoded as float32 with dimensions (X, Y, Z)
• For segmentations should be encoded as binary masks with dimensions (X, Y, Z)

For segmentations: What do you mean by "encoded as binary masks"? that the array should be converted to an np.bool_?
This gives the following error with nifti (for me at least):
HeaderDataError: data dtype "<class 'numpy.bool_'>" not supported

Would for instance int8 be OK instead?
And of course in practice the label file will contain ones and zeros alone (I think this is what you mean by "binary encoding").

AB#5937

[ant0nsc]

Hi @davidkvcs , sorry that this is not clear. In our code, we use np.uint8 for masks. And you're right, it should be zeros and ones alone. I'll try to clarify in the docs.

[davidkvcs]

thanks, sounds good!