Channels for Cellpose

[Lem-P]

I have been using a pretrained Cellpose model to resegment my Xenium data. But I am experiencing mixed results while migrating to Sopa. I am trying to understand why. One thing that I noticed, is that in Sopa, we give the channels as characters and "a nuclear staining, and then a cytoplasm staining" https://github.com/gustaveroussy/sopa/blob/master/workflow/config/example_commented.yaml
While Cellpose uses channel numbers as input and in the reverse order [cytoplasm, nucleus]. I have been digging into Sopa's code and I don't see where the channels' names are converted into number and if they are inverted before giving them to model.eval()

[quentinblampey]

Hi @Lem-P, thanks for reporting, it seems to be an error/misleading comment: indeed, we keep the same order when providing the channels to cellpose. I.e., you need to provide [cytoplasm, nucleus] in the "normal cellpose order". Thanks for noticing, I updated the comment!

Do you have better results when using the correct order?

[Lem-P]

Yes, it is definitely better! It would probably be interesting to have it in the doc for API usage https://gustaveroussy.github.io/sopa/api/segmentation/

Thank you very much for the help.

[quentinblampey]

Nice, thanks for your feedback, and sorry for the confusion...I'll update the doc for the API usage