search

CGLab-GIST / target-aware-denoising

BSD 3-Clause "New" or "Revised" License
6 stars 1 forks source link

Question on drjit backend about denoiser #3

Closed Start1er closed 2 weeks ago

Start1er commented 2 weeks ago

Hi: I try to use your offitical denoiser implementation in mitsuba, but the for rendering time cost without denoising to with denoising is from 0.61s to 3.61s, I think this caused by drjit to pytorch and convert it to drjit backend was high cost in time. So I'm here to ask would you have the plan to make the denoiser methods like Cross_Bilateral or simple_regression to be pure coded in drjit backend, this will save time. If not, would you like to tell me the feasibility of my request, because I want to do it by myself, could it work as my wish? If yes, I will try in drjit coded version in my project.

JeongminGu commented 2 weeks ago

Hi,

I currently do not have plans to implement our denoiser in other versions. Additionally, you might consider exploring Drjit documentation and code examples to see if you can implement the denoisers yourself.

Best, Jeongmin Gu