Sampling masks for validation data?

It would be desirable to be able to reproduce and try to improve upon the “validation” results in Tables 4&6 in the fastMRI paper, during development.
I believe the sampling masks for those tables were generated randomly using the function in the code repo. But one does not know what was the state of the random number generator when those masks were made. Any chance that you could either share the validation sampling masks or in some other way make the validation results exactly reproducible?
Thanks!

Hi Jeff
For table 4, you can run models/cs/run_bart_val.py script which already has a random number seed set. So you should be able to reproduce the results.

For the U-Net results in table 6, we had taken care to set the random number seed in the training script (models/unet/train_unet.py), but it looks like we did not set the random number seed in the models/unet/run_net.py script. Unfortunately, this means that it will not be possible to reproduce the numbers exactly, but you should get results that are very close.

Please do note that, when creating those tables, we evaluated the entire validation data for each acceleration factor.