RawGlobalIllumination * DiffuseFilter do not match GI elements.

[flat]

VRayMax 2.30.01 is used.

This scene, SSS2 material apply to teapot.
http://oakcorp.jp/temp/test_ss2comp_max2012.zip

VRayGlobalIllumination elements and (VRayRawGlobalIllumination * DiffuseFilter) do not match.
fringe seen to the surroundings of teapot.

Of course, it is composite at 32 bits. (Nuke and Pdplayer checked)

Where is a cause?

[joconnell]

It's a problem with the mathematics of compositing and how the raw passes are made. Other people have had more luck by taking the global illumination pass and dividing it by the diffuse filter to make their own Raw GI pass which has no edge issues.

[Richb]

So it isn't just me then, I would love a solution to this problem but it looks like I will just have to continue manually cloning out the edges, not ideal by any means but it seems that people have been looking for a solution to this for years and nothing has developed

[joconnell]

Do the above and you'll get correct edges. I think Vlado's working on it at the minute but in the interim it seems most people either use the above as in this discussion ( http://www.chaosgroup.com/forums/vbu...thout-fringing ) or alternatively don't bother with the raw passes. I asked a few other artists who'd tend to use a lot of nuke on their renders and the majority of them seem to get by with the standard GI, light, ref and spec passes and keep working on the 3d render for longer rather than trying to make big changes that'd need the raw passes. Their though is to keep nuke for the things that only nuke can do and let the 3d be as close as possible in the render rather than trying to fix it after.

[Richb]

Thanks for the response Jo but changes to product colour are vital to our company's workflow and we cant be rendering out 30+ seperate images when only a tweak to the diffuse is necessary. Whether we comp with the raw passes straight from vray or extract them from the vraylight and vraygi there is still very obvious fringing. It's the diffuse which is vital for us but we cant be adjusting lighting and gi at the same time otherwise we might as well be using the beauty pass and not compositing at all

I've posted in every thread I can found relating to this problem and there is no solution, I keep reading that people have had success (no fringing) when extracting raws from the light and gi but this just gives a different kind of fringing, hardly a solution

[joconnell]

I totally understand that, and you're right that using the rawlighting that comes from the vray frame buffer is unusable. In the thread above though, if you divide the normal global illumination pass by the diffuse filter pass you'll take away the diffuse colour contribution and be left with only what the light values of the GI added which is the same as the RawGI pass, except this time it won't have the edge issue. Admittedly you have to make your own RawGI and Raw ref passes using this method, so if you've got to do some kind of configurator you'll still be able to do your RawGI x colour corrected Diffuse filter pass. The lovely folks of Aardman in the thread linked above have a diagram of how it all works!

[Richb]

Sorry Jo maybe I didnt make my reply very clear, I am doing exactly what Aardman are doing, but it simply doesn't work, I am dividing the GI by the diffuse which gives me the RawGI, but there is still fringing

Here is the difference...


Fringing when extracting the raw's as Aardman are doing


Notice the clear fringe

Thanks for the speedy response by the way, much appreciated

ps - there are no alpha's anywhere which might need unpremultiplying/premultiplying

[joconnell]

Hmm - I'm grasping at straws here but I think I read in a thread earlier that if you use SSS2 materials, they also get included in the GI pass as well as the sss2 pass, so you might be getting a contribution from the SSS2 in your GI that isn't being taken into account in your Diffuse filter pass - if you try the same thing with a normal vray material do you get the same issue?

Just had a quick look at the scene you've uploaded - is that the one you're testing with at the minute?

[Richb]

No SSS2 mats being used, all vray standard mats, linear workflow, full float, no AA (have tried AA as well (all filters)), Universal Settings (Have also tried IM/LC), unclamped (tried clamping too) and no sub-pixel mapping (you guessed it - I've tried checking it too)

I felt so close to removing the fringing, this is the last remaining spanner in our machine, everything else we do workflow wise feels pretty bullet proof, but the compositing just lets it down. We end up painting out edge after edge after edge and one of our projects at the minute is likely to result in around 500 images being produced, all of which will need paintovers once they are comped

Headscratcher? Headbanger!

Thank you though Jo you are the first to try and help

[joconnell]

Bugger, I'm a bit stumped by this - I'll have to sit down and look at your scene later but it sounds like you've exhausted most options - I'm not sure if vray for Maya has included any different maths about how the raw passes are derived :/

Just had a quick look at the scene you've uploaded - is that the one you're testing with at the minute? This one doesn't have a gamma workflow in place and there's an vray fast sss 2 material on the teapot? I'll have a fiddle hopefully tomorrow with the dividing and see what results I get over here.

[Richb]

Thanks Jo, hopefully Vlado can shed some light as to whether the maths behind raw passes has been or will be tweaked or if there is a solution which doesn't involve maths

In the meantime - we have edges to paint out

[flat]

sorry.　I had overlooked the following posts.
http://www.chaosgroup.com/forums/vbu...183#post434183

Without using rawGlobalIllumination, a theory uses ( globalIllumination/diffuseFilter) .


It solved.
Thank you!

[raymarcher]

Here some tips, I am not sure if 100 percent correct but take a look,

Things to keep in mind:
-use linear color mapping
-avoid multiply
-do not clamp render elements
-export in full float 32 bit
-in Nuke, use scanline based exr files instead of bucket based exr's, vray renders exr's as bucket format
(you can convert bucket to scanline with nuke output node, or render vrimg and convertr vrimg to scanline exr with vrayimg2exr tool)

Here are two ways to comp your passes, the simpler way with a beauty pass, and a more complex way with you direct and indirect lighting pass.

math beauty:
bck
overlay beauty
+ ( (rfl flt x (rfl / rfl filter) )
+ ( (rfr flt x (rfr / rfr filter) )
+ spc

math full:
bck
overlay ( (clr x (prm lght / clr) ) + (clr x scnd lght / clr) )
+ ( (rfl flt x (rfl / rfl filter) )
+ ( (rfr flt x (rfr / rfr filter) )
+ spc

[Richb]

Just had a quick look at the scene you've uploaded - is that the one you're testing with at the minute? This one doesn't have a gamma workflow in place and there's an vray fast sss 2 material on the teapot? I'll have a fiddle hopefully tomorrow with the dividing and see what results I get over here.

That's not my scene Jo sorry, the teapot scene belongs to FLAT (thread starter)

Here is my scene if you fancy having a play, I'm still to find any convincing argument that the fringing can be solved

http://www.anthony-fraine.com/wp-con...ting_Scene.zip

[Richb]

Quick and cheery update, the 'aardman method' as I am now calling it does work as people have been suggesting. The test scene I was using just happened to be an exception (if anybody wants specifics then I'll be more than happy to divulge)

Since moving on from the test scene and going over some old projects which had mega fringing issues, I have had a 100% success rate (the fringing is dead)

Just to confirm for anyone who hasn't been reading this thread or the countless others on this subject, the workflow for comping elements to fully match the rgb render (to the pixel) is:

VrayLighting/DiffuseFilter = Proper RawLight
VrayGI/DiffuseFilter = Proper Raw GI

Proper Raw GI + Proper Raw Lighting
*
diffuseFilter
+
Reflection
+
Refraction
+
specular
+
any other elements
=
beautifully adjustable comp which matches the RGB!