`LION.CTtools.ct_utils.forward_projection()` does not work with CPU

[ShuvraneelMitra]

Hi @AnderBiguri
Our last merge seems to have been an oversight by both of us: whenever we have torch.cuda.<any_method>() the initialization demands GPU support and hence still throws the same error on my laptop. So I believe we should change it back to my original proposal, device = torch.device('cuda' if torch.cuda.is_available() else 'cpu'). I am currently doing my bachelor's thesis in CT reconstruction and would like to prototype some torch models on my laptop before moving them to the GPU server in my uni. Thanks!

[AnderBiguri]

@ShuvraneelMitra I see. Unfortunately this will be a much bigger issue then, as torch.cuda.get_current_device() is used all over the place in LION, not just where we introduced it. Mostly because any of the code in LION would be unreasonably long to run in CPU anyways, but I see your point.

The correct line then would be:

device = torch.device(torch.cuda.current_device() if torch.cuda.is_available() else 'cpu')

anyway. But there are several places all over LION to change this at. If you find all of the cases, feel free to make another PR. Currently I am on holidays so I won't be able to fix it until september.

[ShuvraneelMitra]

Hey @AnderBiguri thanks for your reply! I guess an alternative would be to mention it upfront in the README that the library is compatible only with GPU-enabled devices. In the meantime I shall try to find all the cases if possible. Till then, let's keep the issue open I guess, in case someone else runs into the same problem. Cheers!