Add warning if SubjectsLoader is not used in PyTorch >= 2.3

[fepegar]

import torch
import torchio as tio

colin = tio.datasets.Colin27()
dataset = tio.SubjectsDataset([colin])

loader = torch.utils.data.DataLoader(dataset, batch_size=1)
next(iter(loader))

[/Users/fernando/git/torchio/src/torchio/data/image.py:188): UserWarning: Using TorchIO images without a SubjectsLoader in PyTorch >= 2.3 might have unexpected consequences. Please replace your PyTorch DataLoader with a SubjectsLoader. See https://github.com/fepegar/torchio/issues/1179 for more context about this problem
  warnings.warn(message, stacklevel=1)
Subject(Keys: ('t1', 'head', 'brain'); images: 3)

loader = tio.SubjectsLoader(dataset, batch_size=1)
next(iter(loader))

{
    't1': {
        'path': ['/Users/fernando/.cache/torchio/mni_colin27_1998_nifti/colin27_t1_tal_lin.nii.gz'],
        'stem': ['colin27_t1_tal_lin'],
        'type': ['intensity'],
        'data': tensor([[[[[210684.4219, 225600.1250, 231532.5156,  ...,      0.0000,
                 0.0000,      0.0000],
           [250516.1406, 225600.1250, 219837.2188,  ...,      0.0000,
                 0.0000,      0.0000],
           [241236.2031, 216532.0469, 222167.8125,  ...,      0.0000,
                 0.0000,      0.0000],
           ...,
           [236576.5156, 239396.7031, 224654.7500,  ...,      0.0000,
                 0.0000,      0.0000],
           [229333.7188, 204913.3281, 226833.9844,  ...,      0.0000,
                 0.0000,      0.0000],
           [187158.8750, 165750.9844, 201131.6875,  ...,      0.0000,
                 0.0000,      0.0000]]]]]),
        'affine': array([[[   1.,    0.,    0.,  -90.],
        [   0.,    1.,    0., -126.],
        [   0.,    0.,    1.,  -72.],
        [   0.,    0.,    0.,    1.]]])
...
        [   0.,    1.,    0., -126.],
        [   0.,    0.,    1.,  -72.],
        [   0.,    0.,    0.,    1.]]])
    }
}

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 84.84%. Comparing base (0314926) to head (8408171).
Report is 34 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1215      +/-   ##
==========================================
- Coverage   87.43%   84.84%   -2.59%     
==========================================
  Files          92       92              
  Lines        5975     5997      +22     
==========================================
- Hits         5224     5088     -136     
- Misses        751      909     +158     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[romainVala]

Hi
yes it is a good idea to give an explicit warning. Just a suggestion to be even more specific :

UserWarning: Using TorchIO images without a SubjectsLoader in PyTorch >= 2.3 might have unexpected consequences. Batch element will be type torchio.Subject whereas is you replace your PyTorch DataLoader with a torchio SubjectsLoader, you will get batch element of type dictionary. See #1179 for more context about this problem

[fepegar]

Thanks, @romainVala. I've updated the warning message, inspired by your suggestion.