RuntimeError - Transform

[HripsimeS]

Dear MONAI community,

I am trying to test the following notebook for 3D Unet segmentation
https://github.com/Project-MONAI/tutorials/blob/master/3d_segmentation/spleen_segmentation_3d_lightning.ipynb

On the stage of "Run the training", the process is interrupting in the middle with following error message "RuntimeError: applying transform <monai.transforms.spatial.dictionary.Spacingd object at 0x0000019CAC2BB970>". I tried to reduce "num_workers" and "batch_size", but it did not help. Can you please provide some solution for this error/bug.

Look forward to hearing from you.

[wyli]

it seems it works fine with the google colab (with a gpu runtime) https://colab.research.google.com/github/Project-MONAI/tutorials/blob/master/3d_segmentation/spleen_segmentation_3d_lightning.ipynb, so it's probably your local environment issue, I'm converting this into a discussion. please follow the issue reporting template with more details so that we might be able to help.

[HripsimeS]

Thank you @wyli. Now I am trying to optimize the same notebook I mentioned above for my dataset, 3D images CT images with their labels. But unfortunately I am getting the following error after execution of "Execute a typical PyTorch training process" part. It seems that the running is having an error on loss_function calculation on line 24 and eventually I received a "RuntimeError: index 3 is out of bounds for dimension 1 with size 2". Do you have some ideas what is going on and how to fix this error? Thanks a lot in advance!

RuntimeError Traceback (most recent call last)
~\AppData\Local\Temp/ipykernel_19120/3283586599.py in
22 optimizer.zero_grad()
23 outputs = model(inputs)
---> 24 loss = loss_function(outputs, labels)
25 loss.backward()
26 optimizer.step()

~\anaconda3\envs\abys\lib\site-packages\torch\nn\modules\module.py in _call_impl(self, *input, **kwargs)
1100 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
1101 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1102 return forward_call(*input, **kwargs)
1103 # Do not call functions when jit is used
1104 full_backward_hooks, non_full_backward_hooks = [], []

~\anaconda3\envs\abys\lib\site-packages\monai\losses\dice.py in forward(self, input, target)
144 warnings.warn("single channel prediction, to_onehot_y=True ignored.")
145 else:
--> 146 target = one_hot(target, num_classes=n_pred_ch)
147
148 if not self.include_background:

~\anaconda3\envs\abys\lib\site-packages\monai\networks\utils.py in one_hot(labels, num_classes, dtype, dim)
89
90 o = torch.zeros(size=sh, dtype=dtype, device=labels.device)
---> 91 labels = o.scatter_(dim=dim, index=labels.long(), value=1)
92
93 return labels

RuntimeError: index 3 is out of bounds for dimension 1 with size 2

[wyli]

looks like the network has out_channels=2 but the input labels have more than 2 classes..