Colon in target label for --merge-regions

[rohanchn]

Hi @mittagessen!

I am working on a segmentation model where I am trying to use --merge-regions. I understand that the format for --merge-regions is target:src.

But my target labels have a : like MainZone:column#1 (SegmOnto) and I am getting an Invalid value for '-mr' error.

Perhaps you have a suggestion to solve this?

[rohanchn]

So, my syntax was wrong as I was using -mr multiple times. This actually is not a problem! Closing this now.

My syntax was indeed wrong, which I corrected, and now it's training. But it still won't pick up the something like MarginTextZone:commentary:MarginTextZone:note because of the colon in the target (and src), which is what I wanted to do in the first place.

I guess this functions handles it, but not sure merely changing the separator would work. https://github.com/mittagessen/kraken/blob/c7562facd4b0f260d2b1dae7e1af1c34bd2dfca8/kraken/ketos/segmentation.py#L40-L53

[mittagessen]

It isn't possible on the command line as the parser of the mapping doesn't support escaping colons. But you can create arbitrary mappings when training with the API (or have dummy identifiers in your data and after training just rename the identifiers in the model metadata).

[rohanchn]

I think I will try training with the API. I find the latter slightly tricky. Thank you!