Monai equivalent to tio.OneOf transform? (ETA 8/13)

[neuronflow]

Is your feature request related to a problem? Please describe.
sometimes it is useful to have multiple augmentation paths

Describe the solution you'd like
I am a fan of tio.OneOf
https://torchio.readthedocs.io/_modules/torchio/transforms/augmentation/composition.html#OneOf

Describe alternatives you've considered
currently I use the tio function
Additional context
I find it so useful that I think it should be part of monai core.

[Nic-Ma]

Hi @neuronflow ,

May I know what's the usage case of this transform? What's the difference between a list of random transforms with different probabilities?

Thanks.

[neuronflow]

@Nic-Ma sorry I missed your response. The difference is that only one of the transforms will be picked from the list. While in the the other case you can end up with 0 to n transforms.

You could use it if you want to run an affine OR elastic transformation on your data but not both.

[Nic-Ma]

Hi @wyli @ericspod @rijobro ,

Sounds interesting, what do you guys think about this feature?
Thanks.

[ericspod]

Imgaug uses these sorts of meta-augments as does other libraries like albumentations. Adding the meta-transforms for doing a sequence of transforms (sometimes) or selecting from a list (someof or oneof) wouldn't be difficult. These would be treated like any other randomized transform and not cached by dataset classes.

[neuronflow]

Yes, OneOf and SomeOf would be highly appreciated :)

[lyndonboone]

Hi @wyli , has anybody started working on this yet? If not, would you like me to try to submit a PR?

[Nic-Ma]

Hi @lyndonboone ,

Thanks for your interest here, welcome your contribution if you like.
I think maybe you can make it as subclass of Compose or another Compose mechanism, @wyli @ericspod @rijobro what do you think?

Thanks.

[wyli]

Sure, sounds great @lyndonboone @Nic-Ma . I haven't looked into the details about how to implement this, but happy to help review/revise the PRs.

[lyndonboone]

@Nic-Ma @wyli great. I'll start to put something together for it.

Thanks!