RandCropByPosNegLabel incompatible with ImageDataset?

[mandelkow]

Describe the bug
When using RandCropByPosNegLabel in conjunction with ImageDataset with both image_files and seg_files (in a segmentation task) one would expect the parameter "label" to default to the seg_file input at runtime. But instead an error is raised: "ValueError: label must be provided." How could this be done?

To Reproduce
Steps to reproduce the behavior:
img_xform = Compose([ EnsureChannelFirst(), RandCropByPosNegLabel( [128]*3, pos=0.9, neg=0.1)])
self.train_ds = ImageDataset(
image_files=[ 'image1.nrrd'],
seg_files= [ 'label1.nrrd' ],
transform=img_xform,
seg_transform=img_xform,
image_only=True, # output image + meta
transform_with_metadata=False, # input image + meta?
)
...

Expected behavior
RandCropByPosNegLabel should use the input "seg_files" from ImageDataset if missing the input "label".

Screenshots
If applicable, add screenshots to help explain your problem.

Environment

Ensuring you use the relevant python executable, please paste the output of:

python -c 'import monai; monai.config.print_debug_info()'

================================
Printing MONAI config...

MONAI version: 1.2.0
Numpy version: 1.23.4
Pytorch version: 2.0.1
MONAI flags: HAS_EXT = False, USE_COMPILED = False, USE_META_DICT = False
MONAI rev id: c33f1ba
MONAI file: /Users/mandelkow/bin/mambaforge/envs/py310/lib/python3.10/site-packages/monai/init.py

Optional dependencies:
Pytorch Ignite version: NOT INSTALLED or UNKNOWN VERSION.
ITK version: NOT INSTALLED or UNKNOWN VERSION.
Nibabel version: 4.0.2
scikit-image version: NOT INSTALLED or UNKNOWN VERSION.
Pillow version: 9.2.0
Tensorboard version: 2.13.0
gdown version: NOT INSTALLED or UNKNOWN VERSION.
TorchVision version: 0.15.2
tqdm version: 4.65.0
lmdb version: NOT INSTALLED or UNKNOWN VERSION.
psutil version: 5.9.4
pandas version: 1.5.1
einops version: 0.6.1
transformers version: NOT INSTALLED or UNKNOWN VERSION.
mlflow version: NOT INSTALLED or UNKNOWN VERSION.
pynrrd version: 0.4.2

For details about installing the optional dependencies, please visit:
https://docs.monai.io/en/latest/installation.html#installing-the-recommended-dependencies

Additional context
According to the docs this is not a problem with dictionary-based datasets.

[KumoLiu]

Hi @mandelkow, thanks for your reporting.
If there are other transforms that rely on other inputs, the ImageDataset may need to be further updated. So it seems it's not general to add such logic inside the ImageDataset. Can't your problem be solved with a normal Dataset and dictionary-based transforms?
Hope it can help you, thanks!

[vmiller987]

This is still an on going issue. Is there no other work around?
I'm honestly wondering if it's an issue with RandCropByPosNegLabel, because I can't even get it to work with other datasets.
I can't seem to find an example of how this transforms works for non dict based.