[Bug]: Output segmentation mask mismatch

[1713mz]

Describe the bug

Hello, I have recently found that in the output images the subplots of image/predicted heatmap and groundtruth/predicted mask/segmentation result does not match, see the following image:

I am pretty sure it was caused by the centercrop operation. After disabling the parameter the results looks reasonable:

Dataset

N/A

Model

N/A

Steps to reproduce the behavior

None

OS information

Not relevant

Expected behavior

None

Screenshots

No response

Pip/GitHub

GitHub

What version/branch did you use?

No response

Configuration YAML

Not relevant

Logs

Not relevant

Code of Conduct

• I agree to follow this project's Code of Conduct

[blaz-r]

Hi, thanks for reporting this. It looks very similar to #1447 so I think there is actually a problem with center cropping.
I think the problem arises because inside visualizer the image is loaded again, since the original was normalized:

anomalib/src/anomalib/post_processing/visualizer.py

Lines 102 to 103 in 9effa29

	height, width = batch["image"].shape[-2:]
	image = read_image(path=batch["image_path"][i], image_size=(height, width))

I belive this needs to be handled differently. Either we also pass center crop data to visualizer, or we pass the normalization data to de-normalize the image.
Looking at the history, it seems like there used to be a denormalization which was removed due to an issue:

anomalib/anomalib/post_processing/visualizer.py

Lines 77 to 78 in 4860abc

	image_result = ImageResult(
	image=Denormalize()(batch["image"][i].cpu()),

@samet-akcay do you maybe remember why that was the case and what should be done in this case since visualizing with center-crop seems to be an issue?

[samet-akcay]

Yeah, I think it was to avoid denormalization if i remember correctly, @djdameln?

@ashwinvaidya17 is currently working on #1530. Maybe we could consider addressing it there, or a separate one?

[Akash-Chokhani]

Hi, I would like to work on this issue.

[blaz-r]

Sure 😊, thanks.

[samet-akcay]

Thanks @Akash-Chokhani !

[samet-akcay]

Please keep an eye on this PR #1706 as it might cause some conflicts to your work.

We plan to merge it by tomorrow

[Akash-Chokhani]

Got it. Thanks for letting me know.

[samet-akcay]

We will be able to address this issue with #2358 and #2334

[watertianyi]

centercrop

@ashwinvaidya17 @samet-akcay How is this solved?