A new noise reduction technique -> any plans on implementing this one in mantra?

   1387   5   2
User Avatar
31 posts
Joined: April 2013
https://www.youtube.com/watch?v=YjjTPV2pXY0 [www.youtube.com]


You guys are always up with new stuff.
User Avatar
1261 posts
Joined: Sept. 2015
I glanced over the actual paper. Don't understand the math and principals behind it all so I can't tell is this paper ‘revealing’ all there is to it, or are there some procedures they are keeping ‘private’ because it's a proprietary endeavor?

Looks like a pursuit by NVidia and their researchers/affiliated universities.

I wonder if it's something NVidia is going to incorporate into their cards at some point where it can be accessed via ‘calls’.

On the other hand, if the paper is ‘revealing’ just about everything there is for someone to take and implement (permissions acquired if needed), yeah I think this could be something very useful to incorporate into Mantra.

If it really does what it is suggesting ( lower render times ), implementing this alongside or as an ‘extra’ feature of Mantra would be great - or even as a new render node tool - “Mantra Jnr.”

But I guess the question too is how much of a difference would the final renders be compared to the ‘standard’ mantra? and aside from render time saving, would it be worth it?

Maybe for hobbyists like myself, it could save me having to doing something like working with Redshift.

Looks like it might take quit a bit of development to incorporate too.

Would be interesting to hear from the folks who have better insights/perspective into these things.
User Avatar
24 posts
Joined: July 2007

A production level implementation is coming in Redshift:

… and Clarisse:

Your text to link here… [blogs.nvidia.com]

Some of the results actually looks better for me. There are imperfections, looks like a painting. With enough sampling and post grain & other usual comp effects can be more photoreal than a raw render, I think. Youtube extremely, but other motion vector based encoders (HDTV platforms, etc) also smash/smudge the content, so for these platforms why waste the rendering power…
User Avatar
56 posts
Joined: July 2006
Guys believe in the end it is only about math.
SideFX is a company that listen but to make a good algorithm with VEX or with a new technology is not simple as we think.
If you watched the last presentation about Houdini 16.5 the man that start the presentation(not remember his name now)told about it, something like “we know about render please wait”.
Today a lot of new render engine are on this new way but lets think a little Houdini have a lot of wonderful news since version 15 probably in future versions we will have good news.
When it no start don't worry we have solution in the market, one more simple thing we have a thread called “thought about Mantra” maybe there is more suitable for this question ;-)
Any way this year is finishing soon let's see Houdini in version 17 :-)
User Avatar
270 posts
Joined: Jan. 2015
Have a look at what renderman is doing with regards to denoising:
https://renderman.pixar.com/whats-new [renderman.pixar.com]
also very nice
User Avatar
1 posts
Joined: March 2018
Someone wrapped the optix Denoiser in a command line application.

https://github.com/DeclanRussell/NvidiaAIDenoiser/releases [github.com]
  • Quick Links