Is it possible to run the monai bundle model inference on CPU

[Bala93]

I am trying to run the inference of https://github.com/Project-MONAI/model-zoo/tree/dev/models/brats_mri_segmentation. I have only 12GB GPU, and it is resulting in out-of-memory error. Is it possible to run the inference in CPU ?

[yiheng-wang-nv]

Hi @Bala93 , could you please change this line:
https://github.com/Project-MONAI/model-zoo/blob/dev/models/brats_mri_segmentation/configs/inference.json#L12
and set the device to cpu?
In addition, I think you can also try to reduce the roi_size to small values, such as (120, 120, 80)

[Bala93]

Hi @yiheng-wang-nv Thanks for your answer. Fyi, I am trying to run the Brats segmentation through Slicer, like this example (https://github.com/Project-MONAI/model-zoo/tree/dev/models/pancreas_ct_dints_segmentation). I tried changing the roi like you suggested, but I was using Task01_Brain tumor (Segmentation decathlon), hence there was a conflict in dimensions. Now, I have downloaded the Brats2019, and each patient has 4 different modalities. May I know what should be the input? Is it like 4 separate files, or single file with stacked volume ? I would really appreciate your help.

[yiheng-wang-nv]

I think both methods are fine. If providing 4 separate files, as for the data list, you just need to put 4 filenames into a list for each sample, and the LoadImaged transform can handle it.