Hi @wyli ,

Thanks for raising the idea.
Could you please help add the doc-string and typical use-cases for the new functions?
I tried to guess what the functions are used and the args mean, but still don't quite understand.

Thanks in advance.

sure, I updated the docstring, it adds hooks to MapTransform.__call__ and InvertibleTransform.inverse, for example the cropper's output will have img_transforms in this case, with img_transforms's value from d['img'].applied_operations:

import torch
from monai.transforms import CenterSpatialCropD

a = torch.zeros((1, 10, 10))
data_dict = {"img": a}
cropper = CenterSpatialCropD("img", roi_size=(7, 7))
output = cropper(data_dict)
print(output.keys())
print(output['img'])
print(output['img_transforms'])

output:

dict_keys(['img', 'img_transforms'])
tensor([[[0., 0., 0., 0., 0., 0., 0.],
         [0., 0., 0., 0., 0., 0., 0.],
         [0., 0., 0., 0., 0., 0., 0.],
         [0., 0., 0., 0., 0., 0., 0.],
         [0., 0., 0., 0., 0., 0., 0.],
         [0., 0., 0., 0., 0., 0., 0.],
         [0., 0., 0., 0., 0., 0., 0.]]])
MetaData
	affine: tensor([[1., 0., 0., 2.],
        [0., 1., 0., 2.],
        [0., 0., 1., 0.],
        [0., 0., 0., 1.]], dtype=torch.float64)

Applied operations
[{class: 'CenterSpatialCrop', extra_info: {'cropped': [2, 1, 2, 1]}, id: 140454425823456, orig_size: (10, 10)}]
Is batch?: False
[{class: 'CenterSpatialCrop', id: 140454425823456, orig_size: (10, 10), extra_info: {'cropped': [2, 1, 2, 1]}}]

Hi @wyli ,

Could you please help add this print(output['img_transforms']) in some unit test to verify?

Thanks in advance.

/black
/integration-test
/black

Could you please help add this print(output['img_transforms']) in some unit test to verify?

Thanks in advance.

thanks, I've updated and now the default is USE_META_DICT=False. please help review again.

/build

Hi @wyli ,

Is there any unit tests or integration tests error after removing all the deepcopy in the inverse functions?

Thanks.

Hi @wyli ,

Is there any unit tests or integration tests error after removing all the deepcopy in the inverse functions?

Thanks.

Don't see any issue, probably it's not needed and is more efficient.