TypeError in transforms notebook

Hi all,

I’m trying to run the integrate_3rd_party_transforms notebook on Google Colab, but it’s not working. First, there’s no line to install MONAI. Second, once installed, I get this error when trying to run the last cell:

TypeError                                 Traceback (most recent call last)
<ipython-input-15-dfd10a517436> in <module>()
      2 check_loader = monai.data.DataLoader(check_ds, batch_size=1)
      3 check_data = monai.utils.misc.first(check_loader)
----> 4 image, label = (check_data['image'][0][0], check_data['label'][0][0])
      5 print(f"image shape: {image.shape}, label shape: {label.shape}")
      6 # plot the slice [:, :, 80]

TypeError: 'NoneType' object is not subscriptable

[Nic-Ma]

Hi @fepegar ,

We may try to move all notebooks to Colab later, and update to use DecathlonDataset. Because we want to do a new refactoring iteration for datasets to support cross-validation soon, so I haven’t changed notebooks.

Thanks.

[fepegar]

Hi @Nic-Ma,

We didn’t test this notebook in Colab so far. Could you please help double confirm whether the Decathlon dataset was loaded correctly?

Oh, sorry. I see now that the path to the data was local. I thought the notebook would be self-contained. I think it would be nice that

1. It was indeed self-contained and one can just press “Run all”. MONAI can download the needed decathlon dataset, is that right?
2. A warning was raised if the input of a transform was empty. Have you considered this?

And whether version of TorchIO can support dict now? I can update the notebook for this part later.

It does, it was added in https://github.com/fepegar/torchio/commit/19ff6e7568ab2c62e579fd0728571760d3e381c4 and it’s available in v0.17.21 (you can try running the code I shared in https://github.com/Project-MONAI/MONAI/issues/851#issuecomment-669139528). I haven’t added support in the transforms mentioned in the commit, message, but it should be fine for most of them.

If the input is a dict, a keys kwarg must be passed. The values of those keys are expected to be 4D tensors with shape (C, D, H, W). I can adapt them further if that would be helpful.