/black

Hi @rijobro @wyli ,

I have updated half of the intensity transforms in this PR based on the proposal in #3087 , could you please help review it first?
I also fixed several minor issues in the intensity transforms in this PR.
If you guys have no concerns, I will try to complete all the other RandomizableTransform ASAP.

Thanks in advance.

Hi @rijobro ,

After offline discussion with @wyli , I will update this PR to align with your proposal(adding a randomize arg).

Thanks.

Ok @Nic-Ma, sounds good. For what it's worth, I wonder if splitting the methods into randomize and compute might be better than my suggestion. I'm not sure. Why did you decide that my way might be better? CC @wyli

Hi @rijobro ,

@wyli thinks that changing to compute() may be too big a modification, I will update this PR to minimize it.
We can consider compute() later if necessary.

Thanks.

Yes @rijobro, because we think having separate compute and __call__ may introduce additional API concepts and add unnecessary complexity for the current use case. We discussed that the simple flag method can already achieve the feature requested.

Ok sounds good. Thanks for the info

Hi @rijobro @wyli ,

In the latest commit, I updated RandGaussianNoise as example to show the proposal:

1. In both randomize() and __call__(), only execute computation when do_transform=True.
2. In __call__(), add a flag randomize to control whether to call self.randomize() function.
3. In dict transform, set prob=1.0 for the array transform in it to make sure always execute.
4. If possible, make sure all the keys of dict transform share the same random factors.

Could you please help review this proposal before I continue to update all the others?

Thanks in advance.

Thanks @Nic-Ma, RandGaussianNoise looks good to me.

Hi @rijobro ,

Do you have any more comments? If not, I will try to update others ASAP.

Thanks.

RandGaussianNoise and RandGaussianNoised look great to me, thanks.

/black

Hi @rijobro @wyli ,

I have updated all the intensity transforms in this PR.
I think this PR is already very big, could you please help review it first?
Then I will try to update other transform categories in another PR soon.

Thanks in advance.

/black

/black

/integration-test

/build

/integration-test