Dataloader from PyTorch supports parallel data sample, but this will block PyCharm debug console.

The solutions was to set the num_workers to 0 during the creation of the DataLoader object.