PatchInferencer utility class

[SerodioJ]

Describe your changes

Utility classes to perform inference in patches using either weighted average or voting methods to combine results.

PatchInferencer is a wrapper class that is initialized with a LightningModule(e.g., trained UNet) and input_size.
When called (forward method), it subdivides the input in patches of size input_size, performs inference using the LightningModule and then reconstructs the output.
It also supports the inference of additional patches so that certain regions have multiple inference results when reconstructing the output (combination can be done with averaging or voting).

Checklist before requesting a review

• I have performed a self-review of my code.
• I have added tests to this functionality.
• All tests are passing.
• I have updated the documentation as needed.
• I have followed the project's coding guidelines.

[GabrielBG0]

@SerodioJ Could you describe how you plan for this to integrate with the rest of the pipeline?

[SerodioJ]

@SerodioJ Could you describe how you plan for this to integrate with the rest of the pipeline?

PatchInferencer is a wrapper class that is initialized with a LightningModule and input_size.
When called (forward method), it subdivides the input in patches of size input_size, performs inference using the LightningModule and then reconstructs the output.
It also supports the inference of additional patches so that certain regions have multiple inference results when reconstructing the output (combination can be done with averaging or voting).

[GabrielBG0]

@SerodioJ @otavioon what you guys think about putting this on a new module celled engines? I don't think it should go to utils (even though I told seriodio to do it)

[otavioon]

LGTM

[otavioon]

I think engines is a good location

[GabrielBG0]

Code has been moved to engines, merging the PR