RandSpatialCropSamples MONAI transformation throws AttributeError: 'dict' object has no attribute 'shape'

[nehagianchandani]

I am trying to crop N samples (each of size 64,64,64) from each image (3d) of my dataset. I am using the RandSpatialCropSamples transformation.

On running my code the following error is thrown:
dict' object has no attribute 'shape' when I call the getitem() function on the Dataset object or when I call the DataLoader function in Monai.

Some important code snippets are added below:

def load_data():
    imgs_list_train = glob.glob(r"C:\Users\nehag\OneDrive\Desktop\sample_data\T1W\*.nii.gz")
    seg_labels_train = glob.glob(r"C:\Users\nehag\OneDrive\Desktop\sample_data\seg\*.nii.gz")
    print(len(imgs_list_train))
    print(len(seg_labels_train))

    #age_labels_train = glob.glob(r"C:\Users\nehag\**ADD IT HERE**")
    age_labels_train= [55, 49]
    
    # Creates dictionary to store nifti files' paths, seg files and age values
    filenames_train = [{"img": x, "seg_label": y, "age_label": z} for (x,y,z) in zip(imgs_list_train, seg_labels_train, age_labels_train)]
    print('filenames train', filenames_train)
    ds_train = monai.data.Dataset(filenames_train, train_transforms)
    print('len', ds_train.__len__())
    print(ds_train)
    print(ds_train.__getitem__(0))     ---------> error as the code doesn't run after this line and doesn't print the getitem output either

    train_loader = DataLoader(ds_train, batch_size=1, shuffle = True, num_workers=1, pin_memory=True, collate_fn=pad_list_data_collate)
    print('getitem', ds_train.__getitem__(0))
    ds_val = monai.data.Dataset(filenames_train, val_transforms)
    val_loader = DataLoader(ds_train, batch_size=1, shuffle=True, num_workers=1, pin_memory=True, collate_fn=pad_list_data_collate)

    return ds_train, train_loader, ds_val, val_loader

My transformations:

train_transforms = Compose(
    [
        LoadImaged(keys=["img", "seg_label"]),
        # AddChanneld("img", "label"),
        EnsureChannelFirstd(keys=["img", "seg_label"]),
        DivisiblePadd(["img", "seg_label"], 16),
        # Orientationd(keys=["image", "label"], axcodes="RAS"),
        RandSpatialCropSamples(
            roi_size=(64,64,64),
            num_samples= 50,
            random_size=False
        ),
        ScaleIntensityd(
            keys=["img", "seg_label"],
            minv=0.0,
            maxv=1.0
        ),
        RandRotated(
            keys=["img", "seg_label"],
            range_x=np.pi / 4,
            prob=0.5,
            keep_size=True
        ),
        #RandSpatialCropSamples(
        #    roi_size=(64,64,64),
        #    num_samples= 50,
        #    random_size=False
        #)
        # RandGibbsNoised(keys=["img", "label"], prob = 0.2)
    ]
)

The error output:

1
1
filenames train [{'img': 'C:\\Users\\nehag\\OneDrive\\Desktop\\sample_data\\T1W\\CC0001_philips_15_55_M.nii.gz', 'seg_label': 'C:\\Users\\nehag\\OneDrive\\Desktop\\sample_data\\seg\\CC0001_philips_15_55_M.nii.gz', 'age_label': 55}]
len 1
<monai.data.dataset.Dataset object at 0x0000018B78CBABF0>
Traceback (most recent call last):
  File "C:\Users\nehag\anaconda3\envs\research\lib\site-packages\monai\transforms\transform.py", line 91, in apply_transform
    return _apply_transform(transform, data, unpack_items)
  File "C:\Users\nehag\anaconda3\envs\research\lib\site-packages\monai\transforms\transform.py", line 55, in _apply_transform
    return transform(parameters)
  File "C:\Users\nehag\anaconda3\envs\research\lib\site-packages\monai\transforms\croppad\array.py", line 744, in __call__
    orig_size = img.shape[1:]
AttributeError: 'dict' object has no attribute 'shape'

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "C:\Users\nehag\anaconda3\envs\research\lib\site-packages\monai\transforms\transform.py", line 91, in apply_transform
    return _apply_transform(transform, data, unpack_items)
  File "C:\Users\nehag\anaconda3\envs\research\lib\site-packages\monai\transforms\transform.py", line 55, in _apply_transform
    return transform(parameters)
  File "C:\Users\nehag\anaconda3\envs\research\lib\site-packages\monai\transforms\compose.py", line 173, in __call__
    input_ = apply_transform(_transform, input_, self.map_items, self.unpack_items, self.log_stats)
  File "C:\Users\nehag\anaconda3\envs\research\lib\site-packages\monai\transforms\transform.py", line 118, in apply_transform
    raise RuntimeError(f"applying transform {transform}") from e
RuntimeError: applying transform <monai.transforms.croppad.array.RandSpatialCropSamples object at 0x0000018B776869B0>

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "C:\Users\nehag\PycharmProjects\multitask\main.py", line 73, in <module>
    ds_train, train_loader, ds_val, val_loader = load_data()
  File "C:\Users\nehag\PycharmProjects\multitask\data_loader.py", line 108, in load_data
    print('getitem', ds_train.__getitem__(0))
  File "C:\Users\nehag\anaconda3\envs\research\lib\site-packages\monai\data\dataset.py", line 105, in __getitem__
    return self._transform(index)
  File "C:\Users\nehag\anaconda3\envs\research\lib\site-packages\monai\data\dataset.py", line 91, in _transform
    return apply_transform(self.transform, data_i) if self.transform is not None else data_i
  File "C:\Users\nehag\anaconda3\envs\research\lib\site-packages\monai\transforms\transform.py", line 118, in apply_transform
    raise RuntimeError(f"applying transform {transform}") from e
RuntimeError: applying transform <monai.transforms.compose.Compose object at 0x0000018B77686EF0>

Process finished with exit code 1

Questions:

1. Why does it say that dict has no attribute shape? I understand that I pass a dict type object to my Dataset function, but isn't that what is supposed to be done? I have tried the same code without the RandSpatialCropSamples transformation and it doesn't throw the error so I am pretty that's the cause.
2. How does the RandSpatialCropSamples know what to apply the transformations on. As per the latest MONAI documentation, no keys parameter has to be passed when composing the transformation. I want to apply the same transformations on my "img" and "seg_label", how do I do that?

I tried the call() function to apply this transformation on a single image and it works as expected.

Thanks.

UPDATE:

I figured out the solution. I was simply supposed to use the dict wrappers of MONAI transforms (transform names ending with 'd') Link