2920 Enhance padding mode for Tensor data

[Nic-Ma]

Fixes #2920 .

Description

This PR enhanced the padding modes and additional args for Tensor data.

Status

Ready

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 -f -u --net --coverage.
• Quick tests passed locally by running ./runtests.sh --quick --unittests.
• In-line docstrings updated.
• Documentation updated, tested make html command in the docs/ folder.

[Nic-Ma]

Hi @wyli and @rijobro ,

I enhanced the padding mode and addtional args in the Pad transform in this draft PR, could you please help review the proposal first?
If you agree with the proposal, I will try to update all the related places in the codebase in this PR.

Thanks in advance.

[rijobro]

It would certainly be great if we could use torch.pad for torch input and numpy.pad for numpy input. I think the reason I made these caveats was because the torch.pad is more limited compared to the numpy version. I suspect this will be reflected by the unit tests, but would be great if you could get it working!

[Nic-Ma]

Hi @rijobro ,

Thanks for your support, I think it's user's responsibility to choose the correct mode for the input data type(numpy or Tensor), we can explicitly show the supported modes of numpy and Tensor in the doc-string. If the passing mode is not compatible, it will raise an error naturally.

Thanks.

[Nic-Ma]

Hi @wyli ,

Do you have any other concerns?
If no, I will try to complete this PR for all the related components following this direction: numpy input -> np.pad, tensor input -> torch.pad, rename np_kwargs -> kwargs.

Thanks.

Thanks.

[wyli]

do we need a suitable default for both numpy and torch inputs?

[Nic-Ma]

do we need a suitable default for both numpy and torch inputs?

Maybe let's change the default mode to constant? @wyli @rijobro What do you guys think?

Thanks.

[wyli]

constant 0 may 'break' the original intensity distribution and the results of mean/max/min would be very different from the original ones...I think 'edge'/'replicate' is a safer choice

[Nic-Ma]

/black

[Nic-Ma]

Hi @wyli , @rijobro ,

I have updated the PR based on the previous discussion, mapping numpy and torch modes to make it compatible.
Could you please help review it?
I didn't update the transforms that don't support Tensor data yet.

Thanks in advance.

[Nic-Ma]

/black

[wyli]

/build

[wyli]

thanks, please see some minor comments inline

[Nic-Ma]

/black