SSS noise sampling RANT, and looking for solutions

[dyei nightmare]

im surprised with the high levels of noise you get with indirect SSS in principled, classic, and pbr shaders,  but im most suprised that sesi does nothing about it...

cleaning the sss-pass with the SSS_quality samplings thing locally in the geometry itself or mantra node doesnt help,  as you know the render times increase a lot, (noise level doesnt help here, dont bring it)   at least 5 or 6 min per frame...   and if you have to render 1000 frames with motion blur at 1920...  this absolutelly is useless...   seems like if you want to render in mantra you have to get used to render for at least two weeks including nights...  

i see indirect lighting-sampling as the greatest weakness of this software, and no, you cant argue otherwise, there is no justification for this.  

as long as sesi doesnt fix it, they are giving free market to other third party render software, like red shift,  without anny need.

i could use redshift, i have it instaled, but why sesi doesnt fix indirect light sampling times?  

im even surprised at this line in the help docs:

 

 "Avoid judging PBR noise based on low-resolution test renders."

 

WHAT???  seriously??  you cant say something like that!!  you are admiting a real problem, and seems like it wont change anyway...  

i see everyone and their dog jumping to redshift because sesi lazyness or something to not fix mantra, and it clearly says whats lacking mantra here...  

 

-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

 

so after this rant,  i wonder for a solution...

are there any way to export a custom matte pass  where you can export only the visible part of indirect lighting with all the direct lighting in matte??   in this way you could export  direct lighting passes with low sampling because you wont have noise and the visible indirect lighting geometry in a separated pass with high sampling values but only those areas where direct lighting is not prseent...  

i mean a way to export direct lighting pass without indirect lighting areas present and another pass where you can export indirect lighting visible areas where direct lighting is not reaching...

so you can export the direct lighting pass with very low settings and the indirect light pass with higher settings but only those visible areas where direct light is not present. 

something like a direct shadow pass negated and multiplicated by the indirect light pass 

im not sure if im being enough clear here, but i think that could be a solution. 

Edited September 1, 2017 by dyei nightmare

[mestela]

The counter argument is mantra is essentially free, and is one project among many at sidefx, and they have limited resources. Other render engines cost money, and those developers can focus on just rendering challenges. Not saying that mantra couldn't be better, I'd love that as much as anyone, but understand sidefx can only do so many things to the software each year, and have to pick their battles. 

If you really need those features and need them to render in a reasonable timeframe, look at other renderers. Choice is good! 

 

[Guest tar]

Just fake it till they make it.  Replicate a quasi screen-space SSS by using a seperate SSS pass, blur it in comp and hold it out with a matte. Another option may be to render as a subpixel layered frame and neatnoise/blur the SSS.  

or just use the physical SSS node...

 

 

SSS_trials.hiplc

Edited September 3, 2017 by tar

[gui]

Perhaps SSS in Arnold is faster than mantra, since it´s Houdini's first implementation of the pathtraced SSS. That said, I rendered last month a fluid sim with refractions, volume and sss and frame times are faster than Arnold, so, perhaps it´s how you track the noise and which tricks (sss, volume, refractions, and so on...) are been used in the scene.