issues running preprocessing: fixed_zero_mean_unit_variance

[pr4deepr]

Hi
I am planning to release a better model than the one I referenced in #52 .
Considering I'm having trouble with the old model in recent deepimagej, I thought I'd fast track this, but I'm running into troubles..
This is a pytorch model and it was running on a previous version of deepimagej, however it was using the macros for pre and post-processing.

As the new version of deepimageJ does not use macros, when I used fixed_zero_mean_unit_variance for preprocessing it throws an error.

[SERVICE-1] {"responseType":"LAUNCH","task":"12250151-c846-49dc-a16c-d861d075e897"}
[WORKER-1] SLF4J: No SLF4J providers were found.
[WORKER-1] SLF4J: Defaulting to no-operation (NOP) logger implementation
[WORKER-1] SLF4J: See https://www.slf4j.org/codes.html#noProviders for further details.
[SERVICE-1] 
[SERVICE-1] <INVALID> 
[SERVICE-1] <INVALID> Loading:     100% |========================================|
[SERVICE-1] {"responseType":"COMPLETION","task":"12250151-c846-49dc-a16c-d861d075e897"}
java.lang.IllegalArgumentException: Invalid method 'fixed_zero_mean_unit_variance' in the specs file. The method does not exist in the Bioimage.io framework.
	at io.bioimage.modelrunner.model.processing.TransformationInstance.findClassInClassPath(TransformationInstance.java:193)
	at io.bioimage.modelrunner.model.processing.TransformationInstance.getTransformationClass(TransformationInstance.java:177)
	at io.bioimage.modelrunner.model.processing.TransformationInstance.build(TransformationInstance.java:166)
	at io.bioimage.modelrunner.model.processing.TransformationInstance.<init>(TransformationInstance.java:63)
	at io.bioimage.modelrunner.model.processing.TransformationInstance.create(TransformationInstance.java:78)
	at io.bioimage.modelrunner.model.processing.Processing.buildPreprocessing(Processing.java:79)
	at io.bioimage.modelrunner.model.processing.Processing.<init>(Processing.java:66)
	at io.bioimage.modelrunner.model.processing.Processing.init(Processing.java:110)
	at io.bioimage.modelrunner.model.Model.runBMZ(Model.java:688)
	at io.bioimage.modelrunner.model.Model.runBMZ(Model.java:614)
	at io.bioimage.modelrunner.model.Model.runBMZ(Model.java:575)
	at deepimagej.Runner.runOnTestImages(Runner.java:146)
	at deepimagej.gui.Gui.lambda$runModelOnTestImage$23(Gui.java:314)
	at java.lang.Thread.run(Thread.java:75

You can download the model from this onedrive link

I think I may need help fixing up the rdf yaml file..

This is the output I expect from the sample image:

---

Something else I saw:

The test-input.npy appears to open as a rotated and flipped image when I click Run on Test Image

The npy file in Python:

When I click on Run on test in deepimagej, the image opens like this:

Cheers
Pradeep

[pr4deepr]

Hi @carlosuc3m
I realized I omitted a normalization step in the preprocessing.

In Python, I preprocess the image by:

• dividing image by 255
• normalize by mean and std, i.e., zero_mean_unit_variance

How do I specify two preprocessing steps in the rdf.yaml?

The dividing image by 255, seems to be the scale_range preprocessing. with min_percentile of 0 and max_percentile of 100.

Cheers
Pradeep

[pr4deepr]

Actually, I've updated and posted on imagesc: https://forum.image.sc/t/bioimageio-preprocessing-question/105581

[pr4deepr]

Closing this here as this is more of a bioimageio spec question than
Posting it in bioimageio spec github page: bioimage-io/spec-bioimage-io#669