Weka probability output


when creating the probability maps during apply classifier of Trainable weka segmentation, the resulting images are not recognized as multichannel. I’m trying to process the output in Cellprofiler and I don’t know how to access different classes for each output. When opening in fiji I need to convert to multichannel image in order to split classes.

Do you know how I can solve this issue?

Thank you in advance.



Can you clarify whether the issue lies on the Fiji/Weka-side of things or Cellprofiler? You can always split the channels in Fiji and save them as single slices before loading them into CellProfiler - no? I guess it’s not clear to me the issue… can you explain it a bit more?

We have written a blog about how to export probability maps from ilastik to CellProfiler. I’m not sure if the same procedure will work for weka probability maps but perhaps taking a look will help: https://blog.cellprofiler.org/2017/01/19/cellprofiler-ilastik-superpowered-segmentation/

Apart from that, CellProfiler is happy to digest multi-channel images or single-channel images, whatever you have got. Just set up NamesAndTypes accordingly to name the images or channels properly, and use ColorToGray if you are importing a color image and want to split into component channels. Hope that gets you started!

1 Like


thanks for your reply, but the issue is in the final format of the Weka probability maps. When I import them to fiji we can not split them:

I will take a look on the blog that @AnneCarpenter proposed.

Thank you.


In case it’s helpful, your issue is because it’s not a multi-channel image that Weka creates, it’s a stack.

Use “Stack to Images” to split the probability maps into separate windows.


1 Like