AttributeError: Can't pickle local object 'SliceDataset.__init__.<locals>.<lambda>' about fastmri HOT 3 OPEN

Hi @King-pand, I got the same error as yours. I found this error caused since pickle cannot take lambda function in a __init__ state of SliceDataset in mri_data.py. I solved this error by replacing self.raw_sample_filter = lambda raw_sample: True with a regular function as follows:

class filter_raw_sample():
    def call(self, raw_sample):
    return True

This class object should be built up ahead of SliceDataset object.

from fastmri.

Hello @King-pand, I haven't been able to reproduce this. It looks like there's an issue with your code in terms of how you set up DistributedDataParallel, perhaps due to a cluster misconfiguration.

from fastmri.

maybe you can set numworkers=0 in dataloader = torch.utils.data.DataLoader(dataset, num_workers=4)

from fastmri.