nnUNetPredictor available in newest nnUNetv2 pip package

[hermancollin]

utilities/packaging/run_nnunet_inference.py

Lines 151 to 154 in de4f1f7

	# ---------------------------------------------------------------
	# OPTION 1: Currently, pip install nnUNetv2 does not have the latest version of nnUNet's inference
	# which is defined in OPTION 2. Hence, this method
	# ---------------------------------------------------------------

Just want to point out that option 2 is now usable with the newest nnUNetv2 pip package (v2.2, released 3 weeks ago)

[valosekj]

Nice! Thanks for letting us know!

TODO:

• update the script

[naga-karthik]

Thanks @hermancollin for opening the issue! I updated the inference code to follow the latest nnunet version

For run_nnunet_inference.py -- refer to commit f66d627
For run_inference_single_subject.py -- refer to commit 7c46dd9

@hermancollin can you please try testing one of these inference scripts and let me know if it works? Thanks!

[hermancollin]

@naga-karthik sure I can test this. Only thing to look out for is that v2.2 requires python >= 3.9 (I know it caused some issues for Mathieu). Maybe this should be specified somewhere.

Also, @valosekj is there a specific reason why the requirements file is a .py file? I thought it was a typo so I renamed the file but if there is a reason for this we can always change it back. Nevertheless, we also need to bump the nnunetv2 version here:

utilities/packaging/run_nnunet_inference_requirements.py

Line 29 in 7c46dd9

nnunetv2==2.1

I pushed a patch for this.

Apart from this, there were problems with the script and I was not able to segment images, both with --path-dataset and --path-images. I fixed some obvious typos in variable names but it still doesn't work on my side. I will take a closer look at this tomorrow.

[valosekj]

Also, @valosekj is there a specific reason why the requirements file is a .py file? I thought it was a typo so I renamed the file but if there is a reason for this we can always change it back.

Good catch! You're right, it is a typo.

[hermancollin]

I fixed some minor stuff in #39. It's possible that run_inference_single_subject is not affected by this change because the user does not have the option to use a folder for segmentation.