Fix onnx export by rewriting GaussianBlur

[ORippler]

Description

• Provide a summary of the modification as well as the issue that has been resolved. List any dependencies that this modification necessitates.

• In PyTorch 1.12, export to onnx fails due failing shape-inference of torch.nn.functional.interpolate, refer Export to onnx fails for some models with Pytorch 1.12 #472.
• Failing exports can be fixed by reimplementing GaussianBlur2d to no longer rely on input shapes for calculation of the kernel size. As a caveat, now the number of channels need to be specified during init

• Fixes Export to onnx fails for some models with Pytorch 1.12 #472

Changes

• Bug fix (non-breaking change which fixes an issue)
• Refactor (non-breaking change which refactors the code base)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

Checklist

• My code follows the pre-commit style and check guidelines of this project.
• I have performed a self-review of my code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing tests pass locally with my changes

[ORippler]

I changed the AnomalyMapGenerator to inherit from nn.Module, should this also be done for all other instances of AnomalyMapGenerator for consistency? Was there a reason I am unaware of for not making all other AnomalyMapGenerator classes instances of nn.Module initially?

[samet-akcay]

I changed the AnomalyMapGenerator to inherit from nn.Module, should this also be done for all other instances of AnomalyMapGenerator for consistency? Was there a reason I am unaware of for not making all other AnomalyMapGenerator classes instances of nn.Module initially?

As far as I'm aware there was no reason. It would be a good idea to inherit nn.Module for every AnomalyMapGenerator for consistency. It could perhaps be another pr since the scope of this pr is gaussian blur

[djdameln]

Thanks for addressing this! I'll approve the PR once we settle on the appropriate location to store the new GaussianBlur class.

[ashwinvaidya17]

Thanks for identifying and fixing this! I am fine with the changes but I'd wait for the consensus from others regarding the location of blur.py before merging.

[ORippler]

Thanks for identifying and fixing this! I am fine with the changes but I'd wait for the consensus from others regarding the location of blur.py before merging.

Moved it to anomalib/models/components/filters, feel free to merge after tests pass