FastSurfer is not compatible with kaggle or other profile-based python configuration of the logging module

[chituan01]

Hi,

I'm using this library in Kaggle. I already specified vox_size but it came up with an error "conform.py: error: argument --vox_size: expected one argument". And as far as I know, even if I don't specify vox_size, there should exist a default value.

This is my code in Kaggle. Before this, I ran with --seg_only successfully.

! FASTSURFER_HOME=$FASTSURFER_HOME FREESURFER_HOME=$FREESURFER_HOME\ $FASTSURFER_HOME/run_fastsurfer.sh --t1 $img \ --sd "{SETUP_DIR}fastsurfer_seg" \ --sid Tutorial \ --py python3 \ --allow_root \ --fs_license "/kaggle/input/license-fastsurfer/license.txt" \ --vox_size 1

The error happened in conform.py. These are few lines at the end of the output:

================== Creating orig and rawavg from input =========================
 
python3 /root/fastsurfer_tutorial/fastsurfer//FastSurferCNN/data_loader/conform.py -i /root/fastsurfer_tutorial/fastsurfer_seg/Tutorial/mri/orig.mgz --check_only --vox_size min --verbose --log /root/fastsurfer_tutorial/fastsurfer_seg/Tutorial/scripts/conform.log
Reading input: /root/fastsurfer_tutorial/fastsurfer_seg/Tutorial/mri/orig.mgz ...
Input /root/fastsurfer_tutorial/fastsurfer_seg/Tutorial/mri/orig.mgz is already conformed! Exiting.

python3 /root/fastsurfer_tutorial/fastsurfer//FastSurferCNN/data_loader/conform.py -i /root/fastsurfer_tutorial/fastsurfer_seg/Tutorial/mri/aparc.DKTatlas+aseg.deep.mgz --check_only --vox_size --dtype any --verbose
usage: 
Script to conform an MRI brain image to UCHAR, RAS orientation, 
and 1mm or minimal isotropic voxels

USAGE:
conform.py  -i <input> -o <output> <options>
OR
conform.py  -i <input> --check_only <options>
Dependencies:
    Python 3.8+
    Numpy
    https://www.numpy.org/
    Nibabel to read and write FreeSurfer data
    https://nipy.org/nibabel/
Original Author: Martin Reuter
Date: Jul-09-2019
conform.py: error: argument --vox_size: expected one argument

Can anyone help me with this issue?
Thank you for reading!

[dkuegler]

You may be using the \ wrong, try removing them

! FASTSURFER_HOME=$FASTSURFER_HOME FREESURFER_HOME=$FREESURFER_HOME\ $FASTSURFER_HOME/run_fastsurfer.sh --t1 $img --sd "{SETUP_DIR}fastsurfer_seg" --sid Tutorial --py python3 --allow_root --fs_license "/kaggle/input/license-fastsurfer/license.txt" --vox_size 1

They are there to escape the newlines.

Other than that, the vox_size argument should work.

But what you are experiencing is related to a different issue.
Please send us your FastSurfer version:
run_fastsurfer.sh --version

[chituan01]

Hi,

Thank you for your response.

1. I tried to remove \ but it seemed not right.
2. The version is 2.4.2+7e53343.

I have attached my notebook here so that you can better understand my issue (conform.py: error: argument—-vox_size: expected one argument).

error-fastsurfer.zip

Thank you.

[dkuegler]

Can you please send your full logfile. It seems to me that your segmentation is sucessful, but the code to extract the voxel size does not work. this would be similar to #668 #668 (comment).

Please also report your OS and your grep version.
grep --version

[chituan01]

Hi, this is my notebook with the full log https://www.kaggle.com/code/tac2103/vox-size-error-fastsurfer.

Grep version: 3.7
OS: Windows 11 Pro, i5-1135G7

However, I am running this on Kaggle with GPU P100 only.

[dkuegler]

So, I did some investigating and the issue is that kaggle seems to auto-configure the logging backend of python to log to the file /tmp/kaggle.log (for reference, Google colab does not seem to do that). But that means FastSurfer is not able to do logging and this not able to use the logging interface to operate properly.

I have not found, where kaggle does this configuration, but it is happening.

So at this point, FastSurfer is not compatible with kaggle. Please use Google Colab without kaggle in the meantime. (Or find where kaggle is interjecting itself into logging or modify FastSurfer such that calls to logging.basicConfig include an additional force=True flag. This would for example be in logging.py --> setup_logging.

[chituan01]

Thank you for your super detailed reply! I will try as you suggested.

[dkuegler]

After internal discussion, we decided to change the way we are determining the voxel size in recon-surf to also address this issue.

Please note, that this will not address the underlying issue in the kaggle configuration, but should address this one specific error you were seeing.

[m-reuter]

should be fixed by #683 avoid reading voxel size information from log file and instead directly from conform image.