Unable to apply transform on Mednist dataset while implementing 2D image registration

[kitu-datta]

Hi ,

I would like to say that on implementing learning based image registration (2D) notebook as mentioned below
https://github.com/Project-MONAI/tutorials/blob/master/2d_registration/registration_mednist.ipynb,
I see I am unable to use transform.
I get "RuntimeError: applying transform <monai.transforms.compose.Compose object at 0x7fe3fc41dd50>" when I am trying to run below cell as shown in screenshot

Where as I was able to load input data successfully

Would appreciate your help here.
Thanks

[wyli]

the current notebook works fine on colab https://colab.research.google.com/github/Project-MONAI/tutorials/blob/master/2d_registration/registration_mednist.ipynb

could you include some info of what you modified and the detailed error messages?

[kitu-datta]

Hi @wyli ,

It is indeed working fine in colab.
However, I am just trying to run the notebook in my EC2 instance. I observed, while executing below cell I get "ERROR: Could not build wheels for monai which use PEP 517 and cannot be installed directly"

it seems it is unable to build MONAI's C++/CUDA extensions. I have attached the notebook here with error logs , in zipped file.

Image_Registration.zip

Hope this helps.
Thanks

[wyli]

I see, you can replace that cell with something like !pip install monai which doesn't use the c++/CUDA ext, but the pipeline should also work

[kitu-datta]

Hi @wyli,

I figured out that I was having problem with the previous kernel.
The notebook works fine in my new environment with !pip install monai.
But is there any alternative way to build extensions.
I tried first pip install pep517 then BUILD_MONAI=1 pip install git+https://github.com/Project-MONAI/MONAI#egg=monai[all] in my new environment, still I get the same error

"Failed to build monai
ERROR: Could not build wheels for monai which use PEP 517 and cannot be installed directly"

If I am not wrong , on building extensions MONAI would have preferred its own extensions instead of the Pytorch/Python native implementations, when calling some relevant python modules

[wyli]

there is an alternative method to install, please follow https://docs.monai.io/en/latest/installation.html#option-2-editable-installation . currently the extensions are experimental, more rigorous evaluations are still needed