1567 add RemoveRepeatedChannel

[leotam]

Signed-off-by: Leo Tam leot@nvidia.com

Fixes #1567 .

Description

Add DeleteChannel inverse operation to RepeatChannel via slicing. I tried squashing the commits which hilarious duplicated them instead. Anyways if it passes review, can fight git more.

Status

Work in progress

Types of changes

• Non-breaking change (fix or new feature that would not break existing functionality).
• Breaking change (fix or new feature that would cause existing functionality to change).
• New tests added to cover the changes.
• Integration tests passed locally by running ./runtests.sh --codeformat --coverage.
• Quick tests passed locally by running ./runtests.sh --quick.
• In-line docstrings updated.
• Documentation updated, tested make html command in the docs/ folder.

[wyli]

thanks, adding the new class to this file will address some of the ci/cd errors https://github.com/Project-MONAI/MONAI/blob/master/monai/transforms/__init__.py#L258

[rijobro]

I find the name slightly confusing. I would expect DeleteChannel to have argument idx: Union[Sequence[int], int], deleting the channels at the specified index/indices. Since this is the inverse of RepeatChannel I think the name should match that more closely, e.g., RemoveRepeatedChannel.

Alternatively, an inverse method could be added to #1514. I've only concentrated on spatial transformations so far, but there's no reason it has to be limited to these use cases (the base class is InvertibleTransform).

[leotam]

That's probably clearer to fit the use case. As for adding the inverse method to your spatial transforms PR, the hesitation would be it bloats the scope of that PR as it's a nontrivial expansion of scope, and makes it harder to test.

Discussing tests, what's this windows-latest CPU test failing? There's no clear error message. @wyli any helpful comments for these two items?

[rijobro]

I think the relevant part of the Windows error is here. It's trying to use pretrained FCN, but failing to download the pretrained network.

[wyli]

agree that this transforms is removing channels according to the 'repeated' pattern, I think RemoveRepeatedChannel is more appropriate...
we could have a separate PR to inverse the repeat channel transform, we need #1514 to provide some generic interfaces

[wyli]

thanks!