Cannot build Deep-Atlas pipeline on my own dataset

[mikami520]

Issue

I followed the deep_atlas tutorial to construct a pipeline for my own dataset. I downsample the dataset to 256x256x128 and the batch size for loading dataset for both registration and segmentation network are 4. However, I got an error below. Then, I tried to decrease the batch size to 1, but still got same errors, could anyone help me solve this issue? I am appreciated your help!

Result

System Info

[mikami520]

This problem was already solved, please see following explanation!!!

[wyli]

thanks for getting back, would you be interested in sharing your solution here? it's potentially useful for other users who arrive here through search engines.

[ebrahimebrahim]

It looks like a CUDA issue of the sort I sometimes mysteriously resolve by rebooting

[mikami520]

@wyli Sure, it is the problem of the version of Torch. Since my CUDA version is 11.4. But I used the MONAI default installation which installed the Torch 1.11.0+cu102. As a result, I uninstalled the old version and re-installed Torch 1.11.0+cu113 which is compatible to my GPU CUDA.

[mikami520]

It looks like a CUDA issue of the sort I sometimes mysteriously resolve by rebooting
it is the problem of the version of Torch. Since my CUDA version is 11.4. But I used the MONAI default installation which installed the Torch 1.11.0+cu102. As a result, I uninstalled the old version and re-installed Torch 1.11.0+cu113 which is compatible to my GPU CUDA.