DL+DiReCT fails on CPU only workstation

[rcruces]

Hello, I'm trying to run DL-DiReCT on a workstation with no CUDA, but 40 CPU Intel(R) Xeon(R) CPU E5-2630 v4 @ 2.20GHz. I think I'm getting the next error because pytorch might be looking for CUDA but I have none.

Is there anyway to work thought this issue?

dl+direct --subject sub-01 --no-cth sub-01_T1w.nii.gz out/

If you are using DL+DiReCT in your research, please cite:
	Rebsamen, M, Rummel, C, Reyes, M, Wiest, R, McKinley, R.
	Direct cortical thickness estimation using deep learning�based anatomy segmentation and cortex parcellation.
	Human Brain Mapping. 2020; 41: 4804-4814. https://doi.org/10.1002/hbm.25159

Input: (0.5, 0.5, 0.5) [RAS]
Resampling
loading checkpoint /data/mica1/01_programs/DL-DiReCT/src/../model/v0_f1.pth
Traceback (most recent call last):
  File "/data/mica1/01_programs/DL-DiReCT/src/../src/DeepSCAN_Anatomy_Newnet_apply.py", line 439, in <module>
    checkpoint = load_checkpoint(model_file, device)
  File "/data/mica1/01_programs/DL-DiReCT/src/../src/DeepSCAN_Anatomy_Newnet_apply.py", line 402, in load_checkpoint
    return torch.load(checkpoint_file, map_location=device)
  File "/home/anaconda3/envs/DL_DiReCT/lib/python3.10/site-packages/torch/serialization.py", line 795, in load
    return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
  File "/home/anaconda3/envs/DL_DiReCT/lib/python3.10/site-packages/torch/serialization.py", line 1012, in _legacy_load
    result = unpickler.load()
  File "/home/anaconda3/envs/DL_DiReCT/lib/python3.10/site-packages/torch/serialization.py", line 958, in persistent_load
    wrap_storage=restore_location(obj, location),
  File "/home/anaconda3/envs/DL_DiReCT/lib/python3.10/site-packages/torch/serialization.py", line 1055, in restore_location
    return default_restore_location(storage, str(map_location))
  File "/home/anaconda3/envs/DL_DiReCT/lib/python3.10/site-packages/torch/serialization.py", line 215, in default_restore_location
    result = fn(storage, location)
  File "/home/anaconda3/envs/DL_DiReCT/lib/python3.10/site-packages/torch/serialization.py", line 185, in _cuda_deserialize
    return torch.UntypedStorage(obj.nbytes(), device=torch.device(location))
RuntimeError: CUDA error: out of memory
ERROR: Segmentation failed

[mrunibe]

It should be possible to run DL+DiReCT on CPU(s) only. If no GPU is detected, the script should fallback to CPU only (takes more time).

Maybe you have a graphic card that is detected as GPU but with very limited amount of memory (check with nvidia-smi). In this case you may try to disable the GPU explicitly with:
export CUDA_VISIBLE_DEVICES=-1

Alternatively, the error above might also because there is not enough memory (RAM) available. You may try:

1. Is your MRI already skull-stripped (brain without skull)? If not, add the --bet option to run brain extraction first
2. If this still fails, you may additionally try the option --lowmem to further reduce memory consumption (see also FAQ)