V-Ray SSS flickering
486 views
Skip to first unread message
Steve Davy
Feb 9, 2014, 4:16:01 PM2/9/14
to Maya Group
Does anyone have any tips for reducing flickering in SSS with V-Ray?
I'm having problems with this -- see the attached SSS pass output from the VFB. The teeth on the left still have pretty bad flickering, even with the Pre-Pass rate set to 2 in the shader, which I think is already pretty high. Render times are also already quite high so going up to 3 isn't really an option.
Everything else in the scene is rendering just fine, so this is purely something to do with SSS calculations.
I'm having problems with this -- see the attached SSS pass output from the VFB. The teeth on the left still have pretty bad flickering, even with the Pre-Pass rate set to 2 in the shader, which I think is already pretty high. Render times are also already quite high so going up to 3 isn't really an option.
Everything else in the scene is rendering just fine, so this is purely something to do with SSS calculations.
Pierre Jasmin
Feb 9, 2014, 9:28:40 PM2/9/14
to maya...@googlegroups.com
Steve,
Not the tip you wanted but,
If it was an urgency, this sort of small flickering can easily be handled in post, the default of our DE:Flicker works just fine on this shot (with Max Change set 100%).
Pierre
Not the tip you wanted but,
If it was an urgency, this sort of small flickering can easily be handled in post, the default of our DE:Flicker works just fine on this shot (with Max Change set 100%).
Pierre
--
You received this message because you are subscribed to the Google Groups "maya_he3d" group.
To unsubscribe from this group and stop receiving emails from it, send an email to maya_he3d+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
Steve Davy
Feb 9, 2014, 10:01:17 PM2/9/14
to Maya Group
Thanks Pierre, might look into it -- it might well be the most cost effective solution for fixing this type of problem which seems to be intrinsic to V-Ray's SSS, and getting rid of it with the renderer seems to require extremely high quality settings....
Date: Sun, 9 Feb 2014 18:28:40 -0800
From: jas...@revisionfx.com
To: maya...@googlegroups.com
Subject: Re: [maya_he3d] V-Ray SSS flickering
Date: Sun, 9 Feb 2014 18:28:40 -0800
From: jas...@revisionfx.com
To: maya...@googlegroups.com
Subject: Re: [maya_he3d] V-Ray SSS flickering
matt estela
Feb 9, 2014, 10:02:35 PM2/9/14
to maya...@googlegroups.com
yep, we just did a shit-ton of deflicker on the lego movie; we found that with pathtracing we could either spend twice as long on the renders and remove 70% of the noise, or spend an extra hour or so in comp to deflicker and remove nearly all of it.
Steve Davy
Feb 9, 2014, 10:08:23 PM2/9/14
to Maya Group
Subject: Re: [maya_he3d] V-Ray SSS flickering
Steve Davy
Feb 9, 2014, 10:29:42 PM2/9/14
to Maya Group
Out of curiosity Matt, were you using V-Ray?
If so, what did you find got rid of the flickering most in-shader -- upping the subsurface sampling or increasing the prepass rate?
I'm not actually even sure what the prepass rate does when raytracing a solid for SSS and the docs are a bit confusing.
If so, what did you find got rid of the flickering most in-shader -- upping the subsurface sampling or increasing the prepass rate?
I'm not actually even sure what the prepass rate does when raytracing a solid for SSS and the docs are a bit confusing.
From: ma...@tokeru.com
Date: Mon, 10 Feb 2014 14:02:35 +1100
Subject: Re: [maya_he3d] V-Ray SSS flickering
To: maya...@googlegroups.com
matt estela
Feb 9, 2014, 11:43:46 PM2/9/14
to maya...@googlegroups.com
Sorry, was a pathtracer implemented in prman with a raytracer back-end developed in-house.
Mike Thomson
Feb 20, 2014, 7:58:47 PM2/20/14
to maya...@googlegroups.com
...also, what did you use to de-noise renders?
matt estela
Feb 20, 2014, 8:10:06 PM2/20/14
to maya...@googlegroups.com
a range of things depending on the type of noise.
-nukes median filter for cheap n nasty
-a motion vector pass based deflicker/blend, a variant of this basically: http://www.nukepedia.com/gizmos/time/deflicker-with-velocity-pass
- neat video for the difficult to remove pathtracer noise, it did most of the heavy lifting. http://www.neatvideo.com/
-projecting held frames back onto geo, but we'd avoid that as it was pretty labour intensive.
matt estela
Feb 20, 2014, 8:17:23 PM2/20/14
to maya...@googlegroups.com
Pierre, have you had a look at the multi-dimensional denoise stuff? we talked about implmenting it at animal, but other things took priority.
Can't find the exact paper, we were discussing, but basically something that'd look at all the available aov's generated (diff, spec, refl, surfacecolour, normals, position, raw light etc), use that to work out what is noise and what is texture detail, where edges are, and steer the denoiser appropriately.
Sort of like these links, except the video one uses info from the monte-carlo sampler which wouldn't be easily output from most renderers.
Pierre Jasmin
Feb 21, 2014, 2:53:22 PM2/21/14
to maya...@googlegroups.com
On 2/20/2014 5:17 PM, matt estela
wrote:
Pierre, have you had a look at the multi-dimensional denoise stuff? we talked about implmenting it at animal, but other things took priority.
Can't find the exact paper, we were discussing, but basically something that'd look at all the available aov's generated (diff, spec, refl, surfacecolour, normals, position, raw light etc), use that to work out what is noise and what is texture detail, where edges are, and steer the denoiser appropriately.
It's interesting, we had a few requests like that over time but
haven't given much thoughts.
Seems there is 2 problems here, the noise and the potential flickering caused by the filling (sparse interpolation) - I also see the reference you provided also adds a third problem as they show the noise changing scale through some lens effect (DOF). Would need to think some more about that :)
Seems the two most useful extra passes to start would be:
1) surface normals tangent to the camera (or derive approximation good enough for filtering from position or Z) and;
2) some fast to render noise-free local illumination RGBA render (assuming the difference between that and the desired global illuminated result is suddenly there is a spot on the wall or the face has lot more shading contrast/details... ).
Theory would be something like you need: (or a variation of that - would need to play)
1) what is sometimes in image processing described as cartoon-texture decomposition - here this would mean you need a residual image that has all the noise in it, a base that has no noise and clean textures, just not all the shading details provided by the GI. Essentially you want to isolate as a separate image a GI pass.
2) and a filter needs a similarity and a distance weight -- distance here meaning if the filter hits a large normal discontinuity it ignores that (it's strenght in a direction weakens as the slope increases) - or if you prefer you want the filter to diffuse following the normals.
I would like to work on this problem, except I have very limited Maya skills. I am pretty much downgraded skill wise to I can load a scene and check someone render settings and render... I would need a number of different data sets to play with - maybe in a few months (busy right now) we could work with AL making this more practical ? (hey! I even worked in the old office shortly 20 years ago now :) )
Cheers (and back to work),
Pierre @ revisionfx.com
Seems there is 2 problems here, the noise and the potential flickering caused by the filling (sparse interpolation) - I also see the reference you provided also adds a third problem as they show the noise changing scale through some lens effect (DOF). Would need to think some more about that :)
Seems the two most useful extra passes to start would be:
1) surface normals tangent to the camera (or derive approximation good enough for filtering from position or Z) and;
2) some fast to render noise-free local illumination RGBA render (assuming the difference between that and the desired global illuminated result is suddenly there is a spot on the wall or the face has lot more shading contrast/details... ).
Theory would be something like you need: (or a variation of that - would need to play)
1) what is sometimes in image processing described as cartoon-texture decomposition - here this would mean you need a residual image that has all the noise in it, a base that has no noise and clean textures, just not all the shading details provided by the GI. Essentially you want to isolate as a separate image a GI pass.
2) and a filter needs a similarity and a distance weight -- distance here meaning if the filter hits a large normal discontinuity it ignores that (it's strenght in a direction weakens as the slope increases) - or if you prefer you want the filter to diffuse following the normals.
I would like to work on this problem, except I have very limited Maya skills. I am pretty much downgraded skill wise to I can load a scene and check someone render settings and render... I would need a number of different data sets to play with - maybe in a few months (busy right now) we could work with AL making this more practical ? (hey! I even worked in the old office shortly 20 years ago now :) )
Cheers (and back to work),
Pierre @ revisionfx.com
matt estela
Feb 21, 2014, 7:06:15 PM2/21/14
to maya...@googlegroups.com
Heh, I'm no longer at Animal (contract finished after Lego), but I can provide a bunch of example render passes. Most renderer these days provide the passes you describe for free, so surface colour sans lighting, direct light, indirect/gi, normals, position etc etc.
Those two examples I have earlier weren't as interesting as a the done we were discussing last year, unfortunately I can't find the link... Argentinian i think, will keep looking. It came with a matlab implementation, so seemed to be a better place to start.
On 2/20/2014 5:17 PM, matt estela wrote:
Pierre, have you had a look at the multi-dimensional denoise stuff? we talked about implmenting it at animal, but other things took priority.
Can't find the exact paper, we were discussing, but basically something that'd look at all the available aov's generated (diff, spec, refl, surfacecolour, normals, position, raw light etc), use that to work out what is noise and what is texture detail, where edges are, and steer the denoiser appropriately.
It's interesting, we had a few requests like that over time but haven't given much thoughts.
Seems there is 2 problems here, the noise and the potential flickering caused by the filling (sparse interpolation) - I also see the reference you provided also adds a third problem as they show the noise changing scale through some lens effect (DOF). Would need to think some more about that :)
Seems the two most useful extra passes to start would be:
1) surface normals tangent to the camera (or derive approximation good enough for filtering from position or Z) and;
2) some fast to render noise-free local illumination RGBA render (assuming the difference between that and the desired global illuminated result is suddenly there is a spot on the wall or the face has lot more shading contrast/details... ).
Theory would be something like you need: (or a variation of that - would need to play)
1) what is sometimes in image processing described as cartoon-texture decomposition - here this would mean you need a residual image that has all the noise in it, a base that has no noise and clean textures, just not all the shading details provided by the GI. Essentially you want to isolate as a separate image a GI pass.
2) and a filter needs a similarity and a distance weight -- distance here meaning if the filter hits a large normal discontinuity it ignores that (it's strenght in a direction weakens as the slope increases) - or if you prefer you want the filter to diffuse following the normals.
I would like to work on this problem, except I have very limited Maya skills. I am pretty much downgraded skill wise to I can load a scene and check someone render settings and render... I would need a number of different data sets to play with - maybe in a few months (busy right now) we could work with AL making this more practical ? (hey! I even worked in the old office shortly 20 years ago now :) )
Cheers (and back to work),
Pierre @ revisionfx.com
Sort of like these links, except the video one uses info from the monte-carlo sampler which wouldn't be easily output from most renderers.
On 21 February 2014 12:10, matt estela <ma...@tokeru.com> wrote:
a range of things depending on the type of noise.
-nukes median filter for cheap n nasty-a motion vector pass based deflicker/blend, a variant of this basically: http://www.nukepedia.com/gizmos/time/deflicker-with-velocity-pass- neat video for the difficult to remove pathtracer noise, it did most of the heavy lifting. http://www.neatvideo.com/-projecting held frames back onto geo, but we'd avoid that as it was pretty labour intensive.
On 21 February 2014 11:58, Mike Thomson <miketh...@gmail.com> wrote:
...also, what did you use to de-noise renders?
On Mon, Feb 10, 2014 at 3:43 PM, matt estela <ma...@tokeru.com> wrote:
Sorry, was a pathtracer implemented in prman with a raytracer back-end developed in-house.
On Monday, February 10, 2014, Steve Davy <stevi...@hotmail.com> wrote:
Out of curiosity Matt, were you using V-Ray?
If so, what did you find got rid of the flickering most in-shader -- upping the subsurface sampling or increasing the prepass rate?
I'm not actually even sure what the prepass rate does when raytracing a solid for
--
Reply all
Reply to author
Forward
0 new messages