Can we get built in denoiser like renderman

[coilbook]

This looks cool and it speeds up rendering by 2-10 times
Would be nice to see this in version 3
http://renderman.pixar.com/view/denoiser

[Goldorak]

This looks cool and it speeds up rendering by 2-10 times
Would be nice to see this in version 3
http://renderman.pixar.com/view/denoiser

I have posted about this in another thread. The short answer is yes, filtering and denoising are interesting options beyond the existing hotpixel filter, and we've explored this extensively, even implementing this in Brigade at one point. It is independent of the renderer in fact. What we have found is that good denoising, even on GPU, can be pretty computationally expensive to do right, and will come with a trade off in terms of blurring high frequency details. May not matter in some cases, but will in others. It is not a silver bullet.

We have other paths that are perhaps even more promising. One of those is much faster light sampling that may get us nearly noise free results in a lot less samples than it takes now (something we've been experimenting with since 1.5). It is going to take more R&D before we can say more about where this fits into our roadmap.

[coilbook]

Thank you!

Looking forward to that new light sampler especially for small lights at night. Hopefully it will also resolve blurred highlights DOF noise too.

[Jolbertoquini]

I have posted about this in another thread...

This the Post...
viewtopic.php?f=24&t=48606

[momade]

talking about denoising. so far i am using standard denoisers available on the market or built into software packages and i am always coming to the point, that i can denoise the R/G/B channels separately. as far as i see this, octane produces "monochromatic" "greyscale" noise. right? i dont know how to put this into better words. but there is no different noise in the color channels. am i wrong?

br.
mo

[MB]

Noise/fireflies seem to be the single biggest weakness in path tracing renderers.
It is the number one complaint in our office from the rendering cognoscenti.

There are two, at least partial solutions to this problem, proposed by octane team members along time ago.
1) Render a larger image say 2x2 times desired final output, or 4x4 times final output, the down size in photoshop
2) Render the image to a lower level of samples 2 or more times, then merge them using the darken operator in photoshop

If these two techniques produce the results people desire, could there not be a rendering method that does this automatically.

e.g. tell octane to generate 4 beauty passes at 2000 spp each and overlay them with a darken method to derive the final beauty pass
e.g. tell octane to render at a given resolution (user would double desired output dimension) and when user deems enough samples have been made, there is an option to resize to half dimensions.

And if it helps, you can elect to combine both methods.

Selling software is as much about internal client politics and the inevitable advocacy groups within, as it is about features, functions and performance.

i can better 'sell' octane if I can explain how it eliminated the need for a second skill set, i.e. Photoshop, and the extra time spent and billed in post processing.
Artistic eyes, the ones we ultimately want are rarely combined with multi product software skills, (let software skills period), keeping it all in a single platform is very valuable.

[coilbook]

Noise/fireflies seem to be the single biggest weakness in path tracing renderers.
It is the number one complaint in our office from the rendering cognoscenti.

There are two, at least partial solutions to this problem, proposed by octane team members along time ago.
1) Render a larger image say 2x2 times desired final output, or 4x4 times final output, the down size in photoshop
2) Render the image to a lower level of samples 2 or more times, then merge them using the darken operator in photoshop

If these two techniques produce the results people desire, could there not be a rendering method that does this automatically.

e.g. tell octane to generate 4 beauty passes at 2000 spp each and overlay them with a darken method to derive the final beauty pass
e.g. tell octane to render at a given resolution (user would double desired output dimension) and when user deems enough samples have been made, there is an option to resize to half dimensions.

And if it helps, you can elect to combine both methods.

Selling software is as much about internal client politics and the inevitable advocacy groups within, as it is about features, functions and performance.
Great ideas but for animations it is too slow I hope one day otoy will add Brigate as a kernel type for animations


i can better 'sell' octane if I can explain how it eliminated the need for a second skill set, i.e. Photoshop, and the extra time spent and billed in post processing.
Artistic eyes, the ones we ultimately want are rarely combined with multi product software skills, (let software skills period), keeping it all in a single platform is very valuable.

Great ideas but for animations it is too slow I hope one day otoy will add Brigate as a kernel type for animations
I am hoping that cloud GPU rendering will be inexpensive. I would pay 20$ per 1 minute of rendered animation at 1080p

[Rikk The Gaijin]

Just found this video from this year's SIGGRAPH

[e3production]

BTW Iray does denoiser/degrain too in a post process. (see attachement)

[Jolbertoquini]

Just found this video from this year's SIGGRAPH

I saw that to is looking really interesting similar to the doc from disney aswell.

http://www.disneyresearch.com/publicati ... mposition/