No option to skip the check when calling inverse method on the transform

[SachidanandAlle]

Example:

https://github.com/Project-MONAI/MONAI/blob/dev/monai/transforms/spatial/array.py#L315-L329

https://github.com/Project-MONAI/MONAI/blob/dev/monai/transforms/inverse.py#L191

    def inverse(self, data: torch.Tensor) -> torch.Tensor:
        transform = self.pop_transform(data)  # how to pass check=False here...?
        # more logic...

        return data

How to pass check=False when we call inverse method on a given transform?
I would like to skip instance ID check while running the Inverse Transform...

Same in case of Dictionary based transform..

[Nic-Ma]

Hi @SachidanandAlle ,

We usually always check the instance ID in transforms to avoid unexpected mistakes.
May I know your use case? Why you want to disable it?

Thanks.

[SachidanandAlle]

The class instances are created at different times.. one cached and one re-created..

[wyli]

currently it's possible to skip the check if the operation id is 'NONE',

MONAI/monai/transforms/inverse.py

Lines 148 to 149 in 31c4209

	# TraceKeys.NONE to skip the id check
	if xform_id == TraceKeys.NONE:

so, before the inverse call, you can set the id if data is a MetaTensor:

from monai.utils.enums import TraceKeys
...

data.applied_operations[-1][TraceKeys.ID] = TraceKeys.NONE
xform.inverse(data)

[SachidanandAlle]

Yes.. currently I am doing the reset of ids after loading the metatensor from cache

[wyli]

ok, I see it's not very convenient for loading from persistent cache, I'll look into this.

[SachidanandAlle]

yes.. this is the typical use case.. that might help to provide you more context/background.

for repeated inference.. monailabel wants to cache the image (MetaTensor) object (after running first few/subset of pre-transforms) and then use it back for next infer calls...