🚀 Add PreProcessor to AnomalyModule

[samet-akcay]

📝 Description

This PR,

• Adds PreProcessor to AnomalyModule
• Removes transforms from datamodule datasets.

✨ Changes

Select what type of change your PR is:

• 🐞 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)
• 📚 Documentation update
• 🔒 Security update

✅ Checklist

Before you submit your pull request, please make sure you have completed the following steps:

• 📋 I have summarized my changes in the CHANGELOG and followed the guidelines for my type of change (skip for minor changes, documentation updates, and test enhancements).
• 📚 I have made the necessary updates to the documentation (if applicable).
• 🧪 I have written tests that support my changes and prove that my fix is effective or my feature works (if applicable).

For more information about code review checklists, see the Code Review Checklist.

[jpcbertoldo]

A sub-feature request that would fit here: (optionally?) keep both the transformed and original image/mask in the batch.

So instead of

            image, gt_mask = self.XXX_transform(batch.image, batch.gt_mask)
            batch.update(image=image, gt_mask=gt_mask)

something like

            batch.update(image_original=batch.image, gt_mask_original=batch.gt_mask)
            image, gt_mask = self.XXX_transform(batch.image, batch.gt_mask)
            batch.update(image=image, gt_mask=gt_mask)

It's quite practical to have these when using the API (i've re-implemented this in my local copy 100 times haha).

[samet-akcay]

A sub-feature request that would fit here: (optionally?) keep both the transformed and original image/mask in the batch.

So instead of

            image, gt_mask = self.XXX_transform(batch.image, batch.gt_mask)
            batch.update(image=image, gt_mask=gt_mask)

something like

            batch.update(image_original=batch.image, gt_mask_original=batch.gt_mask)
            image, gt_mask = self.XXX_transform(batch.image, batch.gt_mask)
            batch.update(image=image, gt_mask=gt_mask)

It's quite practical to have these when using the API (i've re-implemented this in my local copy 100 times haha).

yeah, the idea is to keep batch.image and batch.gt_mask original outside the model. It is not working that way though :)

[jpcbertoldo]

yeah, the idea is to keep batch.image and batch.gt_mask original outside the model

exactly, makes sense : )

but it's also useful to be able to access the transformed one (eg. when using augmentations)

it is not working that way though :)

didnt get this. cause it's not backcompatible?

[samet-akcay]

yeah, the idea is to keep batch.image and batch.gt_mask original outside the model

exactly, makes sense : )

but it's also useful to be able to access the transformed one (eg. when using augmentations)

it is not working that way though :)

didnt get this. cause it's not backcompatible?

oh I meant, it is currently not working, I need to fix it :)

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[codecov]

Codecov Report

Attention: Patch coverage is 88.59316% with 30 lines in your changes missing coverage. Please review.

Please upload report for BASE (feature/v2@95115f9). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
src/anomalib/pre_processing/utils/transform.py	85.50%	10 Missing ⚠️
src/anomalib/pre_processing/pre_processing.py	86.76%	9 Missing ⚠️
.../anomalib/models/components/base/anomaly_module.py	83.33%	3 Missing ⚠️
src/anomalib/models/image/uflow/lightning_model.py	90.00%	2 Missing ⚠️
...rc/anomalib/models/image/csflow/lightning_model.py	80.00%	1 Missing ⚠️
...malib/models/image/efficient_ad/lightning_model.py	91.66%	1 Missing ⚠️
.../anomalib/models/image/fastflow/lightning_model.py	80.00%	1 Missing ⚠️
.../anomalib/models/image/ganomaly/lightning_model.py	80.00%	1 Missing ⚠️
...dels/image/reverse_distillation/lightning_model.py	80.00%	1 Missing ⚠️
...c/anomalib/models/image/winclip/lightning_model.py	87.50%	1 Missing ⚠️

Additional details and impacted files

@@              Coverage Diff              @@
##             feature/v2    #2358   +/-   ##
=============================================
  Coverage              ?   78.14%           
=============================================
  Files                 ?      297           
  Lines                 ?    12990           
  Branches              ?        0           
=============================================
  Hits                  ?    10151           
  Misses                ?     2839           
  Partials              ?        0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[djdameln]

Thanks, feels like we're getting there in terms of design. I have one major concern regarding the configuration of the image size, and several smaller comments.

[ashwinvaidya17]

Thanks. I have a few minor comments

[ashwinvaidya17]

Minor comment, but can we move this to a separate method?

[samet-akcay]

which one would you prefer? _init_pre_processor, _resolve_pre_processor, _handle_pre_processor or _setup_pre_processor

[ashwinvaidya17]

Should we add a check to ascertain whether train_transform is present? Models like VlmAD might not have train_transforms passed to them. I feel it should pick up val or pred transform is train is not available.

[ashwinvaidya17]

How can we ensure that pre_processor callback is called before the other callbacks? Like, is metrics callback dependent on pre-processing first?

[samet-akcay]

In the base model, we will need to ensure the list of callbacks, I guess. For the child classes that inherits this one, we could have something like this:

def configure_callbacks(self) -> Sequence[Callback]:
    """Configure callbacks with parent callbacks preserved."""
    # Get parent callbacks first
    parent_callbacks = super().configure_callbacks()
    
    # Add child-specific callbacks
    callbacks = [
        *parent_callbacks,      # Parent callbacks first
        MyCustomCallback(),     # Then child callbacks
        AnotherCallback(),
    ]
    return callbacks