Brain Dataset Support - Githubissues

tullie commented 4 years ago

This PR adds support for the upcoming brain dataset. Specifically it adds:

Handling of arbitrary sized kspace and target images (in dataloaders and UNet model)
Adding equispaced masks, which can be chosen with a command line argument
Hard coding best regularization weights for each brain sequence type when using TV.

anuroopsriram commented 4 years ago

Added some inline comments. Incorporate them as you see fit.

Additional general comments:

Update the models/cs/README.md file explaining how the run commands differ for brain MRIs and knee MRIs (--mask-type equispaced for brains and --mask-type random for knees).
Similarly, update the models/unet/README.md and models/zero_filled/README.md
Since you renamed MaskFunc to RandomMaskFunc, please also change all usages of MaskFunc accordingly. For example, fastMRI_tutorial.ipynb and data/README.md.

tullie commented 4 years ago

Going to go ahead and merge because I want this in before the blog post tomorrow. Feel free to comment more and i'll fix it up later if needed.

facebookresearch / fastMRI