This is the basic template, but updation in the spec file, we need to look if the display function and categories are performing as expected. As of now, I have kept the previously commented code also (in case that helps to refactor the current version)

Some Notes (might be helpful for PR review)

1. Plugin’s category (like Image Processing, Object Processing, etc.) are defined after looking at the types of inputs and outputs defined from the config. These types (image, measurement, array, file or executable) are normalized to core types—image, object, or measurement—and matched against a predefined sparse matrix to classify the plugin
   1a. But in my understanding cellpose, stardist etc. falls under Object Processing whilst dynamically running this it would be Image Processing. So, feel free to correct the mapping here

2. Currently, the Docker mount path is hardcoded as /data in the generated plugin. To make this more flexible, we could introduce a configurable flag like docker_mount_path so users can customize it as needed. Or we can document that all CellProfiler plugins generated by Bilayers expect /data to be the mounted directory inside the container, and users should ensure their volume mounts follow this convention

3. Current plugin generation is based on CellProfiler’s segmentation template, which defines ImageSegmentation as the hardcoded superclass. Hence, all generated plugins currently inherit from ImageSegmentation, regardless of the actual algorithm type

4. One more thing to verify, type mappings here

Points 2 & 3 may be revisited in future PRs to avoid overloading the current ones

In Object Processing modules (eg. classical segmentation), the template currently assigns LabelSubscriber to inputs (here) even when the pipeline is providing an image (e.g., "DNA").
fix - it should be simple ImageSubscriber?

For the output — here
--> the current implementation is breaking because the generated output is ImageName. This needs to be fixed, as my implementation fails at this point

I was thinking to resolve this comment

I think we want to do this based on subtype of label (or not), right? Because a segmenter might make an image AND an object, for example

The label subscriber might at least partially be an us problem - I notice that in at least Cellpose and Classical Segmentation, we have "labeled" as one of the input types, which it shouldn't be - you shouldn't be segmenting on labels.

I don't fully understand your second question?

Ahh I see! For the classical segmentation algorithm, the issue was that the template generated LabelSubscriber/LabelName in the inputs/outputs. This broke the pipeline because CellProfiler was expecting an image, but a label was provided instead. This should now be resolved if labeled is not passed in the config

I havent tested - but is run function breaking in saving files now?

Sorry, would you plz delete this line in your next commit @bethac07? I missed this whilst merging main into cp_plus_stardist

I havent tested - but is run function breaking in saving files now?

Of the 5 possible plugins, 3 fail to load (instanseg, cellpose, and gaussian) due to errors in writing the config - not great, but not terrible for a first test. Classical and Stardist load but don't work (but I've worked out what needs to be changed for at least Classical, and Stardist needs some of the same changes). I'll plug away at the fixes!

All modules now load in CellProfiler