First of all thanks for the amazing job to implement ICLabel in python!!! 
I read on Github, that in mne-icalabel there should be improvements of ICLabel in the form of other models. What models are there? Can I already use them? Where can I find the documentation? Thanks!
Hello,
You can’t find anything as for now there is nothing else than the port of ICLabel.
We plan on both developing new models, and benchmarking old and new models on a larger variety of datasets, e.g. different electrode counts, different manufacturers, … This part is still a work in progress.
Best,
Mathieu
Thank you for the reply! Will the larger datasets be published?
Most likely. Note that I wrote ‘a larger variety’. I think for now the main objective is to assess how performant the existing models are on different types of recording. Thus, we aim for a couple of recordings (small datasets) per ‘type of recording’, which can be e.g. different manufacturers, different conditions, different electrode counts, dry vs saline vs gel base electrodes, …
Sorry for the misreading.
And yes, further testing sounds reasonable!
Also in the pape it was written that there were quite a lot of training data, but the testing set was actually very small - with only about 130 ICs marked by several experts. More variety on this is also needed.
Looking forward to the new testing results