iLastik Classifier Not Working

Hello:

I have generated a classifier in ilastik that works perfectly in ilastik but once exported gives different results during “classify pixels” in cell profiler. I would equate the difference between the two results to the cell profiler result setting a higher “threshold level” than the result that is shown in ilastik (ie. more of the objects are classified as background than are supposed to be). As per previous posts to this board, I have generated the classifier using the ilastik that comes with cell profiler and I am not using any large features (ie. all of my features in ilastik are medium or smaller). Any ideas? It is not like there is a lot to change since all you really are doing up to that point in cell profiler is loading the image and then classifying the pixels. Thanks for any help provided.

Cheers,
JTM

Jason T Maynes, PhD/MD
Staff Anaesthesiologist
Principal Investigator, Molecular Structure and Function
Hospital for Sick Children
Toronto, Canada

Hi Jason,

Sorry for the late reply! Would you be willing to post some sample images, your pipeline and the ilastik classifier, so we can take a look.

Regards,
-Mark

It turns out it was an image format issue. Unfortunately the images have to be input in the same format as what the ilastik classifier was trained on - if you train it on png images, the ClassifyPixels has to be run on png images, not tiff or any other format. I used png to save space for the classifier but now have to use png files for everything. Not a big deal but took a little bit to figure out.

Cheers,
JTM

Good to hear that you got it figured out! We’ll include this tidbit in our doucmentaton for future use.
-Mark