error when DataLoader's num_workers is not 0

[mengjiexu]

when I use:
train_data = gluon.data.DataLoader(
MyImageFolderDataset(label_len).transform_first(transform_train),
batch_size=batch_size,num_workers=4)
It shows error:
Can't pickle local object 'Dataset.transform_first..base_fn'

when I use:
train_data = gluon.data.DataLoader(
MyImageFolderDataset(label_len).transform_first(transform_train),
batch_size=batch_size,num_workers=0)
It is ok.

[zachgk]

@mxnet-label-bot add [Bug, Python, Gluon, Data-loading]

Thank you for submitting the issue! I'm labeling it so the MXNet community members can help resolve it.

[intgogo]

Same problem in windows server 2016.

[gaomingnudt]

I got the same problem on Win10
mxnet is latest version
and the same code work fine on mac

[abhinavs95]

@mengjiexu looks like #14541 is a duplicate issue of this. Acc @zhreshold comment there the issue is fixed in 1.4.0 and master