OTOY Forums

Hi Guys,

Any news for denoiser? I saw chaos group released a denoiser https://www.youtube.com/watch?v=z-Vu5UcMuF4 is interesting the idea to have a pass to recognize the depth for mask any Idea to generate such pass?

Best,
JO

Some updates on this in the presentations this week at GTC

Goldorak wrote:Some updates on this in the presentations this week at GTC

Awesome news ^^ would be a video, available on line?

Goldorak wrote:Some updates on this in the presentations this week at GTC

at this talk ?
Tuesday, May 9, 10:00 AM - 10:25 AM
https://gputechconf2017.smarteventsclou ... _ID=115931

or this one ?
Wednesday, May 10, 4:30 PM - 4:55 PM
https://gputechconf2017.smarteventsclou ... _ID=114216

Both, as there was some overlap, but the first talk went a little more into the AI R&D work for Octane.

We're going to finish our own AI/DL denoiser R&D Octane imager module (which will be non-destructive - like tone mapping/hotpixel) and then look at the NVIDIA siggraph paper implementaiton to see if there are further improvements we can add from there (as long as they are unencumbered for use on non-NVIDIA hardware).

We did confirm the NV system works with any unbiased MC renderer (they say it is not really designed for biased renders which was interesting - makes sense as optimal biased denoiser would likely be different algorithm). It needs about 1000 sample images to be trained on a particular kernel, and does not need AOVs. This is in line with our own results so far.

Separately, in the GTC talks we showed some recent R&D test work on a highly biased real time (i.e. not final render quality) viewport denoiser being tested in OctaneImager ICEFX module This is targetting instant (60-120+ fps) denosing of RT reflecitons, relighting, AO and more for live biased/baked/light field video compositing as well as game engine/VR/AR playback.

I haven't stopped thinking about this feature since I saw it was coming on Facebook. My mind is stuck on a loop of dreaming about when it is released. ....So excited!

any road map for this module?

Not yet. We're running more training tests right now. I have seen some worst case de-noising samples (i.e. no training for similar scenes) and it still looks pretty good. Definitely going to be useful.

Sounds good, can't wait

Goldorak wrote:Not yet. We're running more training tests right now. I have seen some worst case de-noising samples (i.e. no training for similar scenes) and it still looks pretty good. Definitely going to be useful.

hopefully it will fix noise in specular reflections that are blurred out (DOF) and noise when light shine through octane volume fog