decollate batch

[rijobro]

decollate a batch of data

#1515

[rijobro]

this requires #1625

[wyli]

thanks! this could also be packaged as a post-processing transform? https://github.com/Project-MONAI/MONAI/blob/master/monai/transforms/post/dictionary.py#L33

[rijobro]

I've added the Decollated as a post transform, good idea.

[Nic-Ma]

Hi @rijobro ,

Thanks for developing this new feature, may I know what's the typical use cases?

[rijobro]

Hi @Nic-Ma of course. After the data loader, the data are collated. Images that were C,H,W,[D] become B,C,H,W,[D]. Similarly metadata are merged (e.g., a list of the transforms applied to each image).

To perform the inverse transforms on images, I want to convert them back from B,C,H,W,[D] and collated metadata to a list containing B elements, each element containing C,H,W,[D] images, and only the metadata associated with that image.

My particular use case is for inverse transforms of a batch of images, but I'm sure this isn't the only use case for this functionality (for example, NiftiSaver.save_batch could be rewritten to use it).

[Nic-Ma]

OK, I see, thanks for your explanation.

[wyli]

not sure why it gives a circular import error in the container https://github.com/Project-MONAI/MONAI/runs/1977893666?check_suite_focus=true#step:3:32 I couldn't replicate the issue locally

[rijobro]

Hm, so all the different CI builds were OK, but the docker one is causing a problem? Is this using master branch (since it's already merged)?