pytorch ImageFolder transform bug

[dglubokov]

🐛 Bug

While using the torchvision.datasets.ImageFolder I get an error:

You have to pass data to augmentations as named arguments, for example: aug(image=image)

This is because the ImageFolder uses:

...
if self.transform is not None:
    sample = self.transform(sample)
...

from torchvision/datasets/folder.py.

Is is correct to create an issue here?

Traceback

KeyError                                  Traceback (most recent call last)
<ipython-input-50-60c394f66b5b> in <module>
----> 1 model_ft = train_model(mobilenet_v3_large, criterion, optimizer_ft, exp_lr_scheduler,
      2                        num_epochs=25)

<ipython-input-38-1a5d4f162548> in train_model(model, criterion, optimizer, scheduler, num_epochs)
     20 
     21             # Iterate over data.
---> 22             for inputs, labels in dataloaders[phase]:
     23                 inputs = inputs.to(device)
     24                 labels = labels.to(device)

~/.local/lib/python3.9/site-packages/torchvision/datasets/folder.py in __getitem__(self, index)
    178         sample = self.loader(path)
    179         if self.transform is not None:
    180             sample = self.transform(sample)
    181         if self.target_transform is not None:
    182             target = self.target_transform(target)

~/.local/lib/python3.9/site-packages/albumentations/core/composition.py in __call__(self, force_apply, *args, **data)
    163     def __call__(self, *args, force_apply=False, **data):
    164         if args:
    165             raise KeyError("You have to pass data to augmentations as named arguments, for example: aug(image=image)")
    166         self._check_args(**data)
    167         assert isinstance(force_apply, (bool, int)), "force_apply must have bool or int type"

KeyError: 'You have to pass data to augmentations as named arguments, for example: aug(image=image)'

Expected behavior

Should work while training.

Environment

• Albumentations version: 0.5.2
• Python version: 3.9
• OS: Linux
• How you installed albumentations (conda, pip, source): pip

[Dipet]

Library works only with named arguments and numpy array. For this reason, it is incompatible with torchvision.
Try to wrap transforms using this class

class Transforms:
    def __init__(self, transforms: A.Compose):
        self.transforms = transforms

    def __call__(self, img, *args, **kwargs):
        return self.transforms(image=np.array(img))