Purpose behind optional imports

[jpcenteno80]

Hi, I am trying to build and operator to convert dicom series to nrrd and was trying to follow the logic behind scripting operators already implemented in the library. In monai_seg_inference_operator.py, numpy is imported at the top of the script, and several numpy functions are used in the code. However, torch is imported using optional_import and versioned to 1.5. Also several torch functions are used throughout the remainder of the code.

I was wondering why optional_import is used in some instances?

Thanks!

[MMelQin]

Thanks for the question.

optional_import is used both in MONAI and MONAI Deploy App SDK, so that the package themselves when built/published do not contain those dependent packages, e.g. an application is built to perform classic image processing from a DICOM study, no need to use PyTorch for inference, hence no need to force PyTorch package be in the App SDK.

[dbericat]

@jpcenteno80 could you please confirm Ming's answer solved your question to close this discussion? Thanks!