nnunetv2.inference crashes on import (No module named 'blosc2') due to buggy acvl_utils release

[joshuacwnewton]

Traceback:

  File "/home/runner/sct_0.0/spinalcordtoolbox/deepseg/inference.py", line 27, in <module>
    import spinalcordtoolbox.deepseg.nnunet as ds_nnunet
  File "/home/runner/sct_0.0/spinalcordtoolbox/deepseg/nnunet.py", line 13, in <module>
    from nnunetv2.inference.predict_from_raw_data import nnUNetPredictor   # noqa: E402
  File "/home/runner/sct_0.0/python/envs/venv_sct/lib/python3.9/site-packages/nnunetv2/inference/predict_from_raw_data.py", line 22, in <module>
    from nnunetv2.inference.data_iterators import PreprocessAdapterFromNpy, preprocessing_iterator_fromfiles, \
  File "/home/runner/sct_0.0/python/envs/venv_sct/lib/python3.9/site-packages/nnunetv2/inference/data_iterators.py", line 12, in <module>
    from nnunetv2.preprocessing.preprocessors.default_preprocessor import DefaultPreprocessor
  File "/home/runner/sct_0.0/python/envs/venv_sct/lib/python3.9/site-packages/nnunetv2/preprocessing/preprocessors/default_preprocessor.py", line 23, in <module>
    from nnunetv2.preprocessing.cropping.cropping import crop_to_nonzero
  File "/home/runner/sct_0.0/python/envs/venv_sct/lib/python3.9/site-packages/nnunetv2/preprocessing/cropping/cropping.py", line 5, in <module>
    from acvl_utils.cropping_and_padding.bounding_boxes import get_bbox_from_mask, crop_to_bbox, bounding_box_to_slice
  File "/home/runner/sct_0.0/python/envs/venv_sct/lib/python3.9/site-packages/acvl_utils/cropping_and_padding/bounding_boxes.py", line 5, in <module>
    import blosc2
ModuleNotFoundError: No module named 'blosc2'

This is a duplicate of MIC-DKFZ/acvl_utils#2, but I'm re-reporting here just in case anyone else runs into this bug and tries to find it by checking nnUNet's issues.

[NathanMolinier]

Same problem !

[nis]

Downgrading acvl_utils to version 0.2 gets rid of the initial error, though I haven't run any real work yet. So there might be other problems.
pip install --upgrade acvl_utils==0.2

[elpequeno]

I did the same thing: downgrading acvl_utils to version 0.2 worked for me. So far all my experiments seem to run.

[ykirchhoff]

Hi everyone,

from the acvl-utils issue it seems like this should be resolved. Please let me know if there are any further issues.

Best,
Yannick

[jamesobutler]

Hi @ykirchhoff, nnUNet specifies support for Python 3.9 (https://github.com/MIC-DKFZ/nnUNet/blob/v2.5.1/pyproject.toml#L3), but dependency acvl_utils quietly introduced a dependency of blosc2 in acvl_utils==0.2.1 as described in this issue, and then when the blosc2 dependency was explicitly stated in acvl_utils==0.2.2 it specified blosc2>=3.0.0b4 which requires Python 3.10 or newer (see here). Therefore nnUNet's specification of acvl-utils>=0.2,<0.3 is still problematic for Python 3.9 users. Unfortunately the comment about 0.3 may bring breaking changes was not sufficient, as a patch release of acvl_utils actually brought breaking changes and a drop in python version.

nnUNet/pyproject.toml

Line 34 in 43349fa

"acvl-utils>=0.2,<0.3", # 0.3 may bring breaking changes. Careful!