S20 - Planar 3D Transfer Learning for End to End Unimodal MRI Unbalanced Data Segmentation

[mrkolarik]

General info

• [x] Conference: ICPR 2020
• [x] Submission date: November 24 2020
• [x] Papers title: On the Implementation of Planar 3D Transfer Learning for End to End Unimodal MRI Unbalanced Data Segmentation
• [x] Source paper: Planar 3D Transfer Learning for End to End Unimodal MRI Unbalanced Data Segmentation
• [x] GitHub repo: https://github.com/mrkolarik/transfer2d3d
• [x] Platform: Ubuntu
• [x] Language: Python 3.7
• [x] Results that can be reproduced: Fig. 5 in the original paper - our prediction = pred_682.png in the ./preds/planar_cross_1 folder after prediction. Table 2 in the original paper is a result of cross-validation, we provide weights to reproduce first round of cross-validation, which can be found in the accompanying paper in Table 2.
• [x] Reproduction main step instructions: Complete instruction in the Github repository

Reviewer feedback

• [ ] Tested platform:
• [ ] Result reproduced:
• [ ] Reviewer comment:

Details Results

• Execution without errors:
  • [ ] without changes
  • [ ] with small changes
  • [ ] with a lot of changes
• Results are obtained:
  • [ ] always
  • [ ] sometimes
  • [ ] never
• Results correspond to the paper's figures/tables:
  • [ ] exactly
  • [ ] with negligible differences
  • [ ] with differences
  • [ ] with a lot of differences

[elnrdfrs]

Hello @mrkolarik, I am the reviewer assigned to your paper and implementation for the reproductible label. I did manage to obtain the results that were reported on Figure 5 and Table 2 of your paper. However, I do have some comments on the instructions provided in the GitHub repository:

• The command you provide for creating the conda environment should come before the instruction to run the nifti_to_png.py script.
• Although the opencv package is present in the environment.txt package plan, I still encounter the 'No module named cv2' problem when I launch evaluate_predictions.py. When I tried installing opencv with conda, I encountered the following message:

  conda install -c conda-forge opencv Collecting package metadata (current_repodata.json): done
  Solving environment: | 
  The environment is inconsistent, please check the package plan carefully
  The following packages are causing the inconsistency:
  ...

• I got around the problem (without actually solving it) by installing opencv-python with pip3 outside the planar3d environment.
• Afterwards, evaluate_predictions.py yielded the expected output.

[mrkolarik]

Hello,

thank you for your feedback. As I have already had many troubles with Conda installation before, I decided to add your workaround solution to the open-cv installation problem to our readme.

With kind regards,

Martin Kolarik