initial unet recon demo commit

[mersad95zd]

Signed-off-by: mersad95zd [email protected]

Fixes # .

Description

This PR contains a demo to perform training/validation with U-Net for accelerated MRI reconstruction.

Status

Ready

Checks

• [x] Notebook runs automatically ./runner [-p <regex_pattern>]

[review-notebook-app[bot]]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[Can-Zhao]

I was wondering if you could:

• Add workflow figure to explain the recon algorithm. Current one is too simple.
• Add training/validation curve, so the users can know whether their performance is normal or not.
• Report inference result in readme, and compare with leaderboard to show that this implementation gives reasonable result.

Thank you!

[Can-Zhao]

Thanks.

• Could you remove the model pt? We will put it in model-zoo in the future, not in the tutorial.
• I was not sure whether this folder should be called 2d_reconstruction, since it works on the 2D slices of 3D volumes. Would it be better to put them in folder reconstruction/MRI_reconstruction/?
• Could you put the training curves in the section Training, and inference results in the section Inference?
• You included the number of volumes for training and validation. Could you also include the number of volumes for inference?

[Can-Zhao]

Looks good to me. @wyli Could you help final check it?