Usage of RandCropbyLabelClassesd--ERROR

[sarvanichinthapalli]

import numpy as np
import torch
from monai.transforms import RandCropByLabelClassesd

Create random 3D image data (e.g., [512, 512, 91])

img_np = np.random.uniform(low=-3024, high=3071, size= (512, 512, 91)).astype(np.float32)

Create label with only 0s and 1s, 1s are sparsely placed

lbl_np = np.zeros((512, 512, 91), dtype=np.uint8)
lbl_np[100:300, 100:200, 45:50] = 1 # Inject class 1 label sparsely

print("Image shape:", img_np.shape)
print("Label shape:", lbl_np.shape)

Wrap in dict as expected by MONAI

data = {
"image": img_np,
"label": lbl_np
}

Define cropping transform

crop_size = [32, 32, 91] # height, width, depth

transform = RandCropByLabelClassesd(
keys=["image", "label"],
label_key="label",
spatial_size=crop_size,
num_classes=2,
num_samples=1
)

Apply transform

result = transform(data)[0].........

THrows error

Image shape: (512, 512, 91)
Label shape: (512, 512, 91)

ValueError Traceback (most recent call last)
/tmp/ipython-input-4063268068.py in <cell line: 0>()
31
32 # Apply transform
---> 33 result = transform(data)
34
35 # Show shapes

6 frames
/usr/local/lib/python3.11/dist-packages/monai/transforms/croppad/dictionary.py in call(self, data, lazy)
1148 def call(self, data: Mapping[Hashable, Any], lazy: bool | None = None) -> list[dict[Hashable, torch.Tensor]]:
1149 d = dict(data)
-> 1150 self.randomize(d.get(self.label_key), d.pop(self.indices_key, None), d.get(self.image_key)) # type: ignore
1151
1152 # initialize returned list with shallow copy to preserve key ordering

/usr/local/lib/python3.11/dist-packages/monai/transforms/croppad/dictionary.py in randomize(self, label, indices, image)
1135 self, label: torch.Tensor, indices: list[NdarrayOrTensor] | None = None, image: torch.Tensor | None = None
1136 ) -> None:
-> 1137 self.cropper.randomize(label=label, indices=indices, image=image)
1138
1139 @LazyTransform.lazy.setter # type: ignore

/usr/local/lib/python3.11/dist-packages/monai/transforms/croppad/array.py in randomize(self, label, indices, image)
1339 if _shape is None:
1340 raise ValueError("label or image must be provided to infer the output spatial shape.")
-> 1341 self.centers = generate_label_classes_crop_centers(
1342 self.spatial_size, self.num_samples, shape, indices, self.ratios, self.R, self.allow_smaller, self.warn
1343 )

/usr/local/lib/python3.11/dist-packages/monai/transforms/utils.py in generate_label_classes_crop_centers(spatial_size, num_samples, label_spatial_shape, indices, ratios, rand_state, allow_smaller, warn)
751 center = unravel_index(indices_to_use[random_int], label_spatial_shape).tolist()
752 # shift center to range of valid centers
--> 753 centers.append(correct_crop_centers(center, spatial_size, label_spatial_shape, allow_smaller))
754
755 return ensure_tuple(centers)

/usr/local/lib/python3.11/dist-packages/monai/transforms/utils.py in correct_crop_centers(centers, spatial_size, label_spatial_shape, allow_smaller)
609
610 """
--> 611 spatial_size = fall_back_tuple(spatial_size, default=label_spatial_shape)
612 if any(np.subtract(label_spatial_shape, spatial_size) < 0):
613 if not allow_smaller:

/usr/local/lib/python3.11/dist-packages/monai/utils/misc.py in fall_back_tuple(user_provided, default, func)
294 """
295 ndim = len(default)
--> 296 user = ensure_tuple_rep(user_provided, ndim)
297 return tuple( # use the default values if user provided is not valid
298 user_c if func(user_c) else default_c for default_c, user_c in zip(default, user)

/usr/local/lib/python3.11/dist-packages/monai/utils/misc.py in ensure_tuple_rep(tup, dim)
220 return tuple(tup)
221
--> 222 raise ValueError(f"Sequence must have length {dim}, got {len(tup)}.")
223
224

ValueError: Sequence must have length 2, got 3.