Page 1 of 1
Noise Filtering from Random Parameters
Posted: Thu Oct 13, 2011 8:38 pm
by Chris
Wouldn't this be great?:
http://www.youtube.com/watch?v=ZQ6TFZE5QEU Would it be possible to implement something like this in Octane?
Cheers,
Chris
Re: Noise Filtering from Random Parameters
Posted: Thu Oct 13, 2011 9:40 pm
by t_3
http://octanerender.com/forum/viewtopic.php?f=9&t=8650
by the way, as this is straight post production, it's just a matter of time till a product exists where you can just run you low spp images/animations through...
Re: Noise Filtering from Random Parameters
Posted: Fri Oct 14, 2011 9:01 am
by telemmaite
This is far far from simple post production/noise removal mate. You need multiple different passes(6 if i recall) which guide the algorithm to get the result.
Re: Noise Filtering from Random Parameters
Posted: Mon Oct 17, 2011 9:42 am
by t_3
telemmaite wrote:This is far far from simple post production/noise removal mate. You need multiple different passes(6 if i recall) which guide the algorithm to get the result.
hm, they showed which 6 different parameters their algorithm needs; but later the guy in the vid says: "here's the noisy input to our algorithm, which took about 15sec. per frame to compute, and after a few minutes our algorihtm produces frames like this..." this sounds like postpro to me (i assume he didn't mean they need minutes for frames that took seconds initially ;). also the demo shown was rendered with luxrender, what implys, that they have implemented it into luxrender already (could be possible of course), or the existing output options are sufficient...
Re: Noise Filtering from Random Parameters
Posted: Mon Oct 17, 2011 1:20 pm
by telemmaite
Well you can say its a post pro as it takes the final 8spp image rendered but it is also not a post pro because it needs the output of 6 other passes.
Screen position,Random parameters,World space coords,Surface normals,Texture value,Sample color... so in someway we can say that assigning a material to the viewport in Octane can be considered as post pro...but in reality there are bunch of stuff that happen after. For me a post pro is linear you move a slider you get the result here however because RPF is a complex algorithm you cannot tweak one of the 6 parameters independently. There is also a timing breakdown it takes 33.7% of the time required from start to finish for rendering while the remaining 66.3% of the time is needed for their algorithm to calculate.
Re: Noise Filtering from Random Parameters
Posted: Mon Oct 17, 2011 9:17 pm
by t_3
telemmaite wrote:Well you can say its a post pro as it takes the final 8spp image rendered but it is also not a post pro because it needs the output of 6 other passes.
Screen position,Random parameters,World space coords,Surface normals,Texture value,Sample color... so in someway we can say that assigning a material to the viewport in Octane can be considered as post pro...but in reality there are bunch of stuff that happen after. For me a post pro is linear you move a slider you get the result here however because RPF is a complex algorithm you cannot tweak one of the 6 parameters independently. There is also a timing breakdown it takes 33.7% of the time required from start to finish for rendering while the remaining 66.3% of the time is needed for their algorithm to calculate.
hm, i consider it beeing a little more post pro using the rendering ouput, compared to manipulating the rederer input (=materials)

if you do post production e.g. in after fx you also often handle misc. different render passes for final compositing.
however the important thing imho is, that it doesn't affect the way the render kernel works (at least how i understand it) and hence should be available for a wide range of renderes, including octane (i mean if lux could be used, it must work with octane too). i believe it won't be that complicated to get those extra passes out - and as this method would save tons of render hours, there will be a quite solid demand if a final product is available...
Re: Noise Filtering from Random Parameters
Posted: Tue Oct 18, 2011 8:50 am
by Scog
A very interesting paper, one that looks like it will have a large impact on rendering in general.
The base algorithm is post production as long as you have the required passes, but it seems aspects of it could also be incorporated into the original rendering algorithm as feedback for importance sampling.
This is very, very promising, and looks like the roadmap to realtime photorealism just got a whole lot shorter.