Closed rdrighetto closed 5 months ago
Hey @rdrighetto ! Haven't looked carefully, any reason for cupy rather than torch? Extra deps are bad 😝
I agree torch would be better since we already use it, but cupy is what lazy me got going. I can try with torch later. Feel free to reject or put the PR on hold.
-- Ricardo Diogo Righetto
Em ter., 3 de out. de 2023 às 16:52, alisterburt @.***> escreveu:
Hey @rdrighetto https://github.com/rdrighetto ! Haven't looked carefully, any reason for cupy rather than torch? Extra deps are bad 😝
— Reply to this email directly, view it on GitHub https://github.com/teamtomo/membrain-seg/pull/37#issuecomment-1745144739, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB2W676W53ANIICZKMVHSLTX5QRCVAVCNFSM6AAAAAA5RCLED2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONBVGE2DINZTHE . You are receiving this because you were mentioned.Message ID: @.***>
Hey @rdrighetto ! Thanks for the PR. I think agree that we should use the GPU to perform such operations when possible. I think we should use torch
though to avoid having to pass arrays between libraries and additional dependencies. I will close this for now, but we can always revive it if we decide to go the cupy route 😄
Deconvolution code now runs on GPU using CuPy. Optionally the old CPU code path can still be used via the
--cpu
flag (e.g. if it runs out of memory on the GPU).