About SubjectLoader and Queue

[LucaLumetti]

Is there an existing issue for this?

• I have searched the existing issues

Problem summary

Hello there,
I recently updated torch from version <2.3 to 2.5 and a new warning appeared in my stdout:
UserWarning: Using TorchIO images without a torchio.SubjectsLoader in PyTorch >= 2.3 might have unexpected consequences, e.g., the collated batches will be instances of torchio.Subject with 5D images. Replace your PyTorch DataLoader with a torchio.SubjectsLoader so that the collated batch becomes a dictionary, as expected. See https://github.com/fepegar/torchio/issues/1179 for more context about this issue.

So as suggested, I changed my DataLoader to SubjectLoader, but the warning did not disappear.
So I delved deeper and found that Queue creates a DataLoader, which causes the message above.

torchio/src/torchio/data/queue.py

Line 359 in 9051dbc

subjects_loader = DataLoader(

My code is nearly the same as the example on the docs page: https://torchio.readthedocs.io/patches/patch_training.html
I have also attached the code below in case I missed something important.

I am not sure if this is something intended or if there is maybe something wrong with the way I used Queue and SubjectLoader in my code. As Queue used directly a DataLoader, it is unclear how I can fix this error message.
When I iterate through the subjectloader, I get a dict as expected and everything still works as before.

I also noticed that in_torch_loader() is not extremely fast (still not a bottleneck for sure in my code), maybe a way to disable it would be helpful?

Thanks!

Code for reproduction

dataset = tio.SubjectsDataset(
    subjects,
    transform=transforms,
)
patch_sampler = tio.data.LabelSampler(
    patch_size=(96, 96, 96),
    label_probabilities=label_probs,
)
patches_queue = tio.Queue(
    subjects_dataset=dataset,
    max_length=15,
    samples_per_volume=3,
    sampler=patch_sampler,
    num_workers=num_workers,
)

patches_loader = tio.SubjectsLoader(
    patches_queue,
    batch_size=batch_size,
    shuffle=True,
    num_workers=0,
)

Actual outcome

Multiple warnings (i guess one for each num_workers):
UserWarning: Using TorchIO images without a torchio.SubjectsLoader in PyTorch >= 2.3 might have unexpected consequences, e.g., the collated batches will be instances of torchio.Subject with 5D images. Replace your PyTorch DataLoader with a torchio.SubjectsLoader so that the collated batch becomes a dictionary, as expected. See https://github.com/fepegar/torchio/issues/1179 for more context about this issue.

Error messages

`UserWarning: Using TorchIO images without a torchio.SubjectsLoader in PyTorch >= 2.3 might have unexpected consequences, e.g., the collated batches will be instances of torchio.Subject with 5D images. Replace your PyTorch DataLoader with a torchio.SubjectsLoader so that the collated batch becomes a dictionary, as expected. See https://github.com/fepegar/torchio/issues/1179 for more context about this issue.`

Expected outcome

I would expect the warning to disappear

System info

Platform:   Linux-5.15.0-117-generic-x86_64-with-glibc2.35
TorchIO:    0.20.3
PyTorch:    2.5.1+cu118
SimpleITK:  2.4.1 (ITK 5.4)
NumPy:      1.26.3
Python:     3.9.13 (main, Aug 25 2022, 23:26:10) 
[GCC 11.2.0]