search

sct-pipeline / contrast-agnostic-softseg-spinalcord

Contrast-agnostic spinal cord segmentation project with softseg
MIT License
4 stars 3 forks source link

Inconsistent list of subject used for training #109

Closed jcohenadad closed 3 months ago

jcohenadad commented 3 months ago

This list of subject is 5 months old, and it seems inconsistent with release 2.4, which includes additional data bases.

jcohenadad commented 3 months ago

Related to this, another, probably stale, instruction from this other README: https://github.com/sct-pipeline/contrast-agnostic-softseg-spinalcord/tree/main?tab=readme-ov-file#42-datalist-creation

naga-karthik commented 3 months ago

Hmm, I wasn't sure if I could update the list of subjects. Some of the datasets that were used for release 2.4 are private and I don't know if I can upload the subjects list on GitHub?

jcohenadad commented 3 months ago

Hmm, I wasn't sure if I could update the list of subjects. Some of the datasets that were used for release 2.4 are private and I don't know if I can upload the subjects list on GitHub?

I'm talking about listing the name of the subjects-- not the data. Essentially, we should be able to reproduce your model (and to be able to do so, we need the list of subjects for train/valid/test)

naga-karthik commented 3 months ago

Ah okay! I misunderstood!

That reminds me -- the main branch is up-to-date because I'm still developing the model (i.e. trying to improve the performance on compressions, although not at the moment, but I will get to it). Because of this all the latest updates in the branch nk/new-model and in this folder you can find the all the information about the datasets and splits used.

naga-karthik commented 3 months ago

I am also unsure what to do with the branch actually. I thought this branch will lead to a new model (ie in a different publication where the model is trained on several other datasets). BUT, it is likely going to be delayed because of the revisions for the original contrast-agnostic MedIA paper

what would you suggest in this case?

jcohenadad commented 3 months ago

That reminds me -- the main branch is up-to-date because I'm still developing the model

I'm not sure I understand the logic here. The main branch is not up to date, see e.g. https://github.com/sct-pipeline/contrast-agnostic-softseg-spinalcord/issues/109#issue-2346694612.

Because of this all the latest updates in the branch nk/new-model and in this folder you can find the all the information about the datasets and splits used.

OK I understand the issue now. You're working on a branch that is more recent than the main branch, and your release was based of your branch nk/new-model instead of from the main branch. This is not the proper git workflow and leads to confusion.

The proper workflow is:

naga-karthik commented 3 months ago

The main branch is not up to date

Apologies! I meant to say "NOT" updated but missed writing NOT 😓

Thank you for the proper workflow! That's exactly what I was missing! I have finally gotten around to merging the massive PR. You can find it the full subjects list from all datasets in the datasplits folder in the main branch itself.