Postprocess IHC segmentation based on high synapse count

[schilling40]

Post-process IHC segmentation based on the synapse per IHC count.
First, IHC instances with a high synapse per IHC count are investigated and replaced with multiple instances from a reference segmentation.
Based on the edited segmentation. the synapse count per IHC is once more calculated and IHC instances without synapses can be omitted.

The current commit works for single crops in tif format but seems to have issues when being applied on data in n5 / ome-zarr format.

WIP

[schilling40]

The current implementation of the script flamingo-tools/scripts/prediction/postprocess_ihc_synapse.py uses the segmentation found at the input path both as the input data and the base for the output. Regions of interest are replaced in the input and stored in the same file. While this avoids copying the data and saving the whole file again, it requires to copying of the input files beforehand and is not very clean. What would be a better way to handle this?

[constantinpape]

This looks good.

The current implementation of the script flamingo-tools/scripts/prediction/postprocess_ihc_synapse.py uses the segmentation found at the input path both as the input data and the base for the output. Regions of interest are replaced in the input and stored in the same file. While this avoids copying the data and saving the whole file again, it requires to copying of the input files beforehand and is not very clean. What would be a better way to handle this?

To understand: the script changes the segmentation in-place, and the copy you mention is needed to have a back-up of the initial version to avoid that it's lost. If I understand this correctly: I think this is ok. You could add a flag --bkp to the script that automatically does the copy, e.g. to a file called <INPUT_FILE>.bkp.