Add fastMRI+ data compatibility to SliceDataset

[mmuckley]

We would like to incorporate functionality for using the fastMRI+ dataset in this GitHub repository. In short, this should:

• Be able to download the csv files from the fastMRI+ GitHub.
  • Include a git hash-based versioning system
  • Parse the csv and link it to samples in SliceDataset. It might be nice if parsing were done by the standard library so we didn't have to add a dependency, but pandas could be acceptable with some justification.

All of this should be feasible by subclassing and extending SliceDataset in mri_data.py. I'm posting the issue for anyone that might want to pick up the project.

[mmuckley]

CC @hansenms

[mattnsoleil]

Hello,

In VarNet, The dataset just loads images with the PIL library and does not convert them into necessary types like list or Tensor in SliceDataset. Consequently, this gives an error like TypeError: batch must contain tensors, numbers, dicts or lists; found object. Is it a solution or an update for this problem?

[mmuckley]

@mattnsoleil that is not related to implementing fastMRI+. Please open as separate issue.

[Gaskell-1206]

Hi everyone, I'm Shengjia or Gaskell. I'm working on this issue. If you have any questions or other concerns, please feel free to talk to me. Thanks!