torch transforms - RandRicianNoise, StdShiftIntensity, RandStdShiftIntensity

[rijobro]

Description

torch transforms - RandRicianNoise, StdShiftIntensity, RandStdShiftIntensity

Status

Ready

Types of changes

• Non-breaking change (fix or new feature that would not break existing functionality).
• New tests added to cover the changes.
• Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.
• Quick tests passed locally by running ./runtests.sh --quick --unittests.

[rijobro]

@wyli I continue to be hampered by the NdarrayTensor problem. Could we come up with a solution that doesn't require # type: ignore all over the place? There's not much point in type checking if we then have to use lots of ignores.

FYI, NdarrayTensor = Union[np.ndarray, torch.Tensor] would work for me. Although I'm sure it would break things elsewhere.

[wyli]

@wyli I continue to be hampered by the NdarrayTensor problem. Could we come up with a solution that doesn't require # type: ignore all over the place? There's not much point in type checking if we then have to use lots of ignores.

FYI, NdarrayTensor = Union[np.ndarray, torch.Tensor] would work for me. Although I'm sure it would break things elsewhere.

for now perhaps in the new feature, we don't include any typing annotations, we can gradually add them later with non-breaking pull requests.

[wyli]

/build

[Nic-Ma]

Looks good to me except for some minor comments.
@wyli Do you have any other comments?

Thanks.

[Nic-Ma]

I think our transforms are in-place operation, no need to deepcopy?

[rijobro]

@Nic-Ma @wyli are we sure about the in-placeness of the transforms? When I deleted the deepcopy, i needed to modify the unit tests for them to pass - 2431dba.

[wyli]

you're right, looks like a bug here, this astype will copy if the datatypes doesn't match, if that happens it's not an in-place operation

MONAI/monai/transforms/intensity/array.py

Line 283 in c39f6fa

img = img.astype(self.dtype)

[rijobro]

@wyli temporarily using NdarrayTensorUnion = Union[np.ndarray, torch.Tensor]. Happy to revert in the future once we find a solution.

[wyli]

/build

[rijobro]

Thanks @Nic-Ma , updated for both those suggestions.

[wyli]

/build