TorchIO_MONAI_PyTorch_Lightning.ipynb issue

[seungyunL]

Hi, I have trouble running TorchIO_MONAI_PyTorch_Lightning.ipynb.

---

data.prepare_data()
data.setup()
print('Training: ', len(data.train_set))
print('Validation: ', len(data.val_set))
print('Test: ', len(data.test_set))

shows

Training: 208
Validation: 52
Test: 130

---

start = datetime.now()
print('Training started at', start)
trainer.fit(model=model, datamodule=data)
print('Training duration:', datetime.now() - start)

shows

LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0,1]

| Name | Type | Params

0 | net | UNet | 122 K
1 | criterion | DiceCELoss | 0

122 K Trainable params
0 Non-trainable params
122 K Total params
0.245 Total estimated model params size (MB)

and following error occurs:

FileNotFoundError: Caught FileNotFoundError in DataLoader worker process 0.
Original Traceback (most recent call last):
File "/home/devin/anaconda3/envs/pytorch_p37/lib/python3.7/site-packages/torch/utils/data/_utils/worker.py", line 287, in _worker_loop
data = fetcher.fetch(index)
File "/home/devin/anaconda3/envs/pytorch_p37/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 52, in fetch
return self.collate_fn(data)
File "/home/devin/anaconda3/envs/pytorch_p37/lib/python3.7/site-packages/pytorch_lightning/utilities/auto_restart.py", line 474, in _capture_metadata_collate
data = default_collate(samples)
File "/home/devin/anaconda3/envs/pytorch_p37/lib/python3.7/site-packages/torch/utils/data/_utils/collate.py", line 157, in default_collate
return elem_type({key: default_collate([d[key] for d in batch]) for key in elem})
File "/home/devin/anaconda3/envs/pytorch_p37/lib/python3.7/site-packages/torch/utils/data/_utils/collate.py", line 157, in
return elem_type({key: default_collate([d[key] for d in batch]) for key in elem})
File "/home/devin/anaconda3/envs/pytorch_p37/lib/python3.7/site-packages/torch/utils/data/_utils/collate.py", line 157, in default_collate
return elem_type({key: default_collate([d[key] for d in batch]) for key in elem})
File "/home/devin/anaconda3/envs/pytorch_p37/lib/python3.7/site-packages/torchio/data/image.py", line 574, in init
super().init(*args, **kwargs)
File "/home/devin/anaconda3/envs/pytorch_p37/lib/python3.7/site-packages/torchio/data/image.py", line 140, in init
self.path = self._parse_path(path)
File "/home/devin/anaconda3/envs/pytorch_p37/lib/python3.7/site-packages/torchio/data/image.py", line 377, in _parse_path
return [self._parse_single_path(p) for p in path]
File "/home/devin/anaconda3/envs/pytorch_p37/lib/python3.7/site-packages/torchio/data/image.py", line 377, in
return [self._parse_single_path(p) for p in path]
File "/home/devin/anaconda3/envs/pytorch_p37/lib/python3.7/site-packages/torchio/data/image.py", line 365, in _parse_single_path
raise FileNotFoundError(f'File not found: "{path}"')
FileNotFoundError: File not found: "data"

Any solutions?

[anafyp]

I just so happen to run into the same issue so if you have found a solution I would love to hear about it.

[KumoLiu]

Hi @anafyp, I think it's the error due to the version of PyTorch, did you try the torch==1.10.2 set in the tutorial?
Thanks!

[anafyp]

Actually, my issue was from a very particular case. I converted all my data to tio.ScalarImages and tio.LabelMaps and then I rsync'ed' them to my computation cluster. once there, when my dataloader was trying to open them since tio works as a lazy loader, it was looking to load the path. even when I set it to ""... so, in the end, I just rsynced the nifties and converted everything on the distant machine... so I didn't solve the issue but found a workaround... This still happened when I tried to image.load() then save with torch...

[KumoLiu]

Seems the behavior is the same in the tutorial. Did you try it with torch==1.10.2, when I using this version, I didn't see the error.