masking_method in Mask class is not saved as argument (preventing applying the inverse transform)

[lchauvin]

Is there an existing issue for this?

• I have searched the existing issues

Problem summary

When trying to get an inverse transform after composing a transform with a Mask(masking_method='mymask'), the masking_method is not saved as argument (like it's the case with ZNormalization for example).
Then, when trying to get the inverse transform, I get:

File "/scratch/02/lchauvin/workspace/miniconda3/envs/VntrcNet/lib/python3.9/site-packages/torchio/data/subject.py", line 212, in apply_inverse_transform
  inverse_transform = self.get_inverse_transform(**kwargs)
File "/scratch/02/lchauvin/workspace/miniconda3/envs/VntrcNet/lib/python3.9/site-packages/torchio/data/subject.py", line 198, in get_inverse_transform
  history_transform = self.get_composed_history(
File "/scratch/02/lchauvin/workspace/miniconda3/envs/VntrcNet/lib/python3.9/site-packages/torchio/data/subject.py", line 176, in get_composed_history
  transforms = self.get_applied_transforms(
File "/scratch/02/lchauvin/workspace/miniconda3/envs/VntrcNet/lib/python3.9/site-packages/torchio/data/subject.py", line 160, in get_applied_transforms
  transform = name_to_transform[transform_name](**arguments)
TypeError: __init__() missing 1 required positional argument: 'masking_method'

Code for reproduction

import torchio as tio

if __name__ == '__main__':
    data = tio.Subject(mask=tio.LabelMap('path_to_mask'),image=tio.ScalarImage('path_to_image'))
transform = tio.Compose([
    tio.ToCanonical(),
    tio.CropOrPad(target_shape=(100,100,100), mask_name='mask'),
    tio.ZNormalization(masking_method='mask'),
    tio.Mask(masking_method='mask'),
    ])

transform_data = transform(data)

transform_data.apply_inverse_transform()

Actual outcome

TypeError: init() missing 1 required positional argument: 'masking_method'

Error messages

Traceback (most recent call last):
  File "/scratch/02/lchauvin/workspace/projects/VntrcNet/predict.py", line 115, in <module>
    main(hp, name=args.name)
  File "/scratch/02/lchauvin/workspace/projects/VntrcNet/predict.py", line 70, in main
    subject_data_inverse = subject_data.apply_inverse_transform()
  File "/scratch/02/lchauvin/workspace/miniconda3/envs/VntrcNet/lib/python3.9/site-packages/torchio/data/subject.py", line 207, in apply_inverse_transform
    inverse_transform = self.get_inverse_transform(**kwargs)
  File "/scratch/02/lchauvin/workspace/miniconda3/envs/VntrcNet/lib/python3.9/site-packages/torchio/data/subject.py", line 193, in get_inverse_transform
    history_transform = self.get_composed_history(
  File "/scratch/02/lchauvin/workspace/miniconda3/envs/VntrcNet/lib/python3.9/site-packages/torchio/data/subject.py", line 171, in get_composed_history
    transforms = self.get_applied_transforms(
  File "/scratch/02/lchauvin/workspace/miniconda3/envs/VntrcNet/lib/python3.9/site-packages/torchio/data/subject.py", line 154, in get_applied_transforms
    transform = name_to_transform[transform_name](**arguments)
TypeError: __init__() missing 1 required positional argument: 'masking_method'

Expected outcome

The inverted transform

System info

No response