'_MultiProcessingDataLoaderIter' object has no attribute 'dataset'

[ffiioonnaa]

Hi，
Thanks for your excellent work! I encountered an error while running the training code
ProcessingTraceback (most recent call last):

  File "traineval.py", line 418, in <module>
    main(args)
  File "traineval.py", line 298, in main
    fig=fig,
  File "/home/zhao/obman_train/mano_train/netscripts/epochpass3d.py", line 71, in epoch_pass
    for batch_idx, (sample) in enumerate(loader):
  File "/home/zhao/obman_train/mano_train/datautils.py", line 18, in __next__
    if isinstance(loader.dataset, Subset):
AttributeError: '_MultiProcessingDataLoaderIter' object has no attribute 'dataset'

and I try to use ' loader._dataset' instead of 'loader.dataset' ,then I met another error

  File "/home/zhao/obman_train/mano_train/networks/branches/manobranch.py", line 181, in forward
    root_palm=root_palm,
  File "/home/zhao/anaconda3/envs/obman_train/lib/python3.7/site-packages/torch/nn/modules/module.py", line 541, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/zhao/anaconda3/envs/obman_train/lib/python3.7/site-packages/manopth/manolayer.py", line 141, in forward
    self.th_hands_mean + th_full_hand_pose
RuntimeError: The size of tensor a (45) must match the size of tensor b (3) at non-singleton dimension 3

[toharl]

pytorch 1.1.0 has solved the issue for me