Richard Brown

@KrisThielemans info RE including ITK. SimpleITK's method looks good but potentially a lot of work to switch over.

It's a shame, our current solution to only install on linux or to not install on windows is to call pip from conda which defeats the purpose of conda: ```yaml...

Hi, a little bump on this. Has anyone had a chance to have a look? I wonder if this potential bug exists in the forward and back projector, but doesn't...

I wonder if there's anything we can test with simulated data. For example, what would we expect if we projected the brainweb data and then NAC reconstructed that?

@casperdcl my original reconstruction and the one in your notebook use projectors where the data is copied between the GPU and CPU. But there's also an OSEM implementation that works...

ah. well good to tick that one off the list at least!

Just reconstructed the same data using STIR/SIRF. Both NACs with norm and randoms, top row is 1 ray, bottom row is 10 rays. 

1, 10, 100 rays: