Closed hermancollin closed 10 months ago
Now I only looked at the _run_nnunetinference.py file. _run_inference_singlesubject.py requires less refactoring. It seems there is only one call to the predict_from_files()
method and the input are more contained.
While looking at the single subject script, I noticed something that could be improved. Some functions are duplicates in _run_nnunetinference and _run_inference_singlesubject. For example, splitext
and add_suffix
are exact copies. That's not a major problem but could become annoying if a bug is fixed on one side but not on the other. I would advise to delete them in one file and import them from the other for easier maintenance.
So the reason the
--path-dataset
arg did not work for me was because theconvert_filenames_to_nnunet_format
was restricted to NIFTI files. However, for--path-images
, there were problems with the code.One of the advantage of the new inference interface is that folders or lists of files can be processed using the same method:
nnUNetPredictor.predict_from_files
supports both. Now, this should work for everyone.Fixes #38