khanlab
AD trained model
yeah its probably not great that we're parsing the model from the name of the URL - probably something we should work on in the future (or here). Either way, I just uploaded Brad's AD model to zenodo so it can be run just like the other models! just wetrun testing now to see how it looks
yeah true at some point we can refactor code to avoid that.. but the easy thing to do right now is move to zenodo as you've done!
made a minor naming change for future proofing.. Also, note that by merging into master we won't have this in the dev-2.0.0 branch -- thinking maybe we just merge into dev-2.0.0 instead..
there are still a few issues with the .tar file - the data from fold_0 seems to be missing, and the folders have a different naming scheme than expected. @Bradley-Karat do you still have the original by any chance? also note that "tempmodel" shouldn't be in the tarred file, it just gets extracted to a folder named that. not sure if you renamed some of the other files/folders, but they should look something like this:
tempmodel/nnUNet/3d_fullres/Task301_ADNI_T1w_successful/Task301_ADNI_T1w_successful__nnUNetPlansv2.1/fold_0/nnUNetTrainer.model.pkl
tempmodel/nnUNet/3d_fullres/Task301_ADNI_T1w_successful/Task301_ADNI_T1w_successful__nnUNetPlansv2.1/fold_0/nnUNetTrainer.model
ah good catch, not sure why fold_0 was missing from that one - I must've uploaded the wrong one. The model with fold_0 can be downloaded here: https://osf.io/kch9v/download
OK the new model should be available, could someone please run a quick test to ensure its publicly accessible and not just to me?
Can alse edit in any additional information on the zenodo page, i tried to make it similar to the neonatal project: https://zenodo.org/uploads/15297857
Do we want to merge this into dev-2.0.0 instead of master? Or we could cherry-pick from dev-2.0.0 and keep this merging into master as is..
i was thinking the latter since there's nothing specific to dev here, and technically the AD model has already been available for a while via this branch & docker build prior to the PR. But i'm fine either way