PDA

View Full Version : Specular indirect



Mats Andersen
06-07-2018, 02:50 AM
Hi Everyone:)


It seems that all the problematic noise in my interior renderings in particular is found in the "specular indirect" pass..

I have increased both antialiasing and reflection samples to unheard values without good results.

Does anyone have a solution to how to either get around or disable "specular indirect"

best regards
Mats

Tobian
06-07-2018, 05:14 AM
https://www.youtube.com/watch?v=72AAwSFx4nA

https://www.youtube.com/watch?v=qeLLGPwiYOs

Try watching these and see if it helps. Without knowing what specifically is causing the specular indirect it's hard to diagnose, but you can disable it by unticking glossy reflections, but then you're basically disabling... reflection.

Mats Andersen
06-08-2018, 02:58 AM
https://www.youtube.com/watch?v=72AAwSFx4nA

https://www.youtube.com/watch?v=qeLLGPwiYOs

Try watching these and see if it helps. Without knowing what specifically is causing the specular indirect it's hard to diagnose, but you can disable it by unticking glossy reflections, but then you're basically disabling... reflection.


Thanks for the tips. I watched these from before and they are helpful videos in many aspects. But still it occurs that most of the noise (fireflies in the worst) in the interiors sems to be laying in the specualr indirect pass.
I understand that it is proapbly a result of too few rays, but since interpolation GI smoothing doesnt work anymore (at least the way it did) this is very limiting. Setting samples to 5000 and antialiasing to 5000 is not a solution duer to the rendering times.

Maybe in future updates there is a way to cheat this pass to become smother? like if it uses the few rays it has and multiplies them driven by an accuracy setting to simulate a larger number of secondary rays than it really has?? Thsi would leave the antialiasing with an easier task of smoothing it.

M

Tobian
06-08-2018, 04:39 AM
You kind of need to post specific examples, because I am having to imagine what your problem is. most 'noise' in any renderer is due to samples being taken across a high contrast region, the smaller and higher contrast the difference the more chance of missing, and therefore fireflies. There's methods to dealing with it depending on the scenario, but 5000 samples seem excessive :D Personally I would debug what it is that is causing the fireflies, and if it's something you can diminish by using tricks, For example, you can use the ray switch to swap out materials with regards to indirect specular only, which might solve it in some cases. Or is this from exterior / sky lighting, in which case you should really be looking into portal lights.

lardbros
06-08-2018, 05:36 AM
5000??? I've not had to go anywhere near that kind of level. 128 for GI rays, or a maximum of 12 for my antialiasing seems to be good for most scenes.

Depending on how rough your metals are, reflection samples can be anything from 4 to 128, but higher than that and you probably need to investigate why there is so much noise going on.

If you can post some examples, or scenes, it would really help.

Mats Andersen
06-08-2018, 06:52 AM
Just joking about the samples hehe.

The example is a large interior space. I want the specular from the distant light to affect the materials, in particualr the balcony fronts where the worst fireflies is.

No matter how I turn the samples of the distant light or reflection samples its still no good.. No HDR, just a regular backdrop and enviroment light set to 1.0 LUX. distant light is set to about 3.0 LX.

Turning off the affect specual on the distant light of course remove the fireflies - but I would not loose the specualar!

Morgan Nilsson
06-08-2018, 07:09 AM
I've noticed that the interpolated GI doesn't update unless you manually rebake it.

Also, do you have portals or environment lights? If that is the case make sure to set your shaders to raytracing only if I recall correctly, can't double check at the moment.

Mats Andersen
06-08-2018, 07:10 AM
by turning reflection samples up over 200 and use 20 samples on AA I kind of get rid of it, but then the animation will never be finsihed...

Kryslin
06-08-2018, 10:30 AM
If you are using an environment light, you will want it to have twice as many samples as your other lights.

lardbros
06-08-2018, 12:05 PM
Thanks for drip feeding some info ;)

Is it possible to see a bit more? Like, what are your GI settings, what are your samples settings? What hdr do you have in there, what are your surface settings? (Specifically in the second tab in the surface editor).
You could probably even get away with turning off glossy reflections for something like those balconies.

Mats Andersen
06-12-2018, 01:02 AM
Thanks for drip feeding some info ;)

Is it possible to see a bit more? Like, what are your GI settings, what are your samples settings? What hdr do you have in there, what are your surface settings? (Specifically in the second tab in the surface editor).
You could probably even get away with turning off glossy reflections for something like those balconies.


Hi

Gi settings 20
No HDR just a distant light and plain backdrop. Enviroment light.
PBR material with glossy reflections.

anyway the noise is not in any way related to light samples and gi settings. It is 100% a problem in the specualr indirect channel. There should be a way to smoothen this or simulate a greater amount of rays in this pass.

I am surprised that not anyone else seem to have this issue. We are 3 guys in the studio here and all our interior renderings are badlty suffering from this problem. To such a degree that we have to revert to 2015 for some scenes.
Again for a still render its possible to crank reflection samples and AA to ridicoulus levels, that gets rid of it to somem degree- allthoug render times is horrendous. For animation its completely hopeless.

Mats

lardbros
06-12-2018, 01:25 AM
Cool, so you're using the environment light. How many samples on your environment light?
Did you check the buffers, and all the noise is in there?

Did you change all your materials to raytrace only? When you use the environment light, you have to turn off the raytrace AND backdrop and change them to raytrace only.

Mats Andersen
06-12-2018, 04:09 AM
Cool, so you're using the environment light. How many samples on your environment light?
Did you check the buffers, and all the noise is in there?

Did you change all your materials to raytrace only? When you use the environment light, you have to turn off the raytrace AND backdrop and change them to raytrace only.


I checked all the buffers and the noise is distinct only in one buffer and thats the "specular indirect"..

All materials is PBR and raytrace only

Samples on enviro light does not affect it. I have also turned off backdrop for reflections.

Tobian
06-12-2018, 06:13 AM
Ok so 3 more things here, because frankly we're guessing based on some very dark images...

1) for Interiors, do not use Environment lights. Use portal lights. The environment light is not optimal for an interior. Put an area light in front of every opening (and you can stretch them to fit several in a row). Set the light to be a portal light, so that it will sample the environment BG, set it's intensity to 1LX, and inverse squared falloff.

2) go through each of the lights in your scene and make sure it does not have a falloff node inside of it, if it does, delete it, set ALL Lights to inverse squared falloff and adjust to suit. I said that in my specular video but you might have skipped :D

3) This is possibly the cause but maybe not... Are all of the materials in there PBR? if you have any standard material that in and of it's self could be the explanation as that's capable of generating 300% reflections without trying, if it's being used and wrong... Standard should be renamed DO NOT USE, because it will cause issue, unless it is very carefully modified. Also just based on all of those bright spots, you should make sure there's some roughness on every surface.

Mats Andersen
06-12-2018, 07:29 AM
Ok so 3 more things here, because frankly we're guessing based on some very dark images...

1) for Interiors, do not use Environment lights. Use portal lights. The environment light is not optimal for an interior. Put an area light in front of every opening (and you can stretch them to fit several in a row). Set the light to be a portal light, so that it will sample the environment BG, set it's intensity to 1LX, and inverse squared falloff.

2) go through each of the lights in your scene and make sure it does not have a falloff node inside of it, if it does, delete it, set ALL Lights to inverse squared falloff and adjust to suit. I said that in my specular video but you might have skipped :D

3) This is possibly the cause but maybe not... Are all of the materials in there PBR? if you have any standard material that in and of it's self could be the explanation as that's capable of generating 300% reflections without trying, if it's being used and wrong... Standard should be renamed DO NOT USE, because it will cause issue, unless it is very carefully modified. Also just based on all of those bright spots, you should make sure there's some roughness on every surface.


Hi

the render I posted is just the specular indirect pass in order to better highlight the concern:-)
It is clrealy a matter of too little samples for this pass.

Also tried portal lights but still the same problem. Still need to go super duper high on the settings to fix the noise.

PBR yes on everything and also the falloff is good.

I will go ahead and make a test scene that is easy to experiment with

M

rustythe1
06-12-2018, 07:47 AM
are you using brute force, you say you have 20 gi, but if using interporlated that setting is meaningless, if using brute force, that's where your problem lies, bring the camera aa back to 4 and reflection samples back to 8 or so, then keep upping the brute force gi (you will probably find 100 to 200 it will start to clean up)

kolby
06-12-2018, 08:17 AM
Turning off the affect specual on the distant light of course remove the fireflies - but I would not loose the specualar!

What is the Angle value of the distant light ?

Tobian
06-12-2018, 09:22 AM
Hmm from this i suspect the main culprit is probably the sun light it's self and that's damn hard to actually fix, other than using some clever node tricks. Work out which objects are causing bright secondary reflections of the sun, and use the ray type switch to swap out for a non glossy material for reflections, which should help kill some of the worst offenders.

lardbros
06-13-2018, 01:49 AM
Ah... Just thought. Was this scene brought over from an old LW scene?

Even if it wasn't, check the lights, falloff nodal stuff. Importing old lights
from 2015 scenes into LW 2018 attempts to recreate the falloff, or no falloff that was used in the old LW.

Basically, go to all of your lights and untick the node stuff, and then make sure the falloff is set to inverse.

Mats Andersen
06-13-2018, 02:48 AM
Thanks for the input guys.

No GI is used in order to isolate the problem.
Distant light has about 0.5 degrees

Ill have to look more.

Again it would be awesome if there in future version of LW was a way to fake/predict added samples in this specular indirect so that very high AA and sample settings could be avoided.
Like interpolation.
GI interpolation in the new lightwave I have stopped using as I dont undertstand it anymore, hehe. Seems to not be as efficient in creating a smooth interpolation as before.The new interpolation smoothness setting only goes half the way.
Only brute force now. :-)

M

Tobian
06-13-2018, 03:15 AM
Yes typically i only use brute force now, as while it's slow it's predictable, and less prone to issues, if slower. I struggle with interpolated too tbh. yes, theoretically brute force can result in noisy indirect specular, it's mostly the reflections of reflections, from very bright (necessarily) light sources, which causes fireflies, which is why i was establishing if you had used old-standard, as that can magnify and make fireflies so much worse!

There's a few paths to solving indirect specular, from using bi-directional path tracing, and VCM, which resolve causicts much better than MC, raytracing, to simply clamping indirect specular, which produces less realistic but normalised results, which are easier to clean. TBH if you look into it every single PBR based renderer which handles indirect specular (and indirect diffuse to an extent) suffers from fireflies, because you cannot multiple importance sample indirect, and either clamp indirect rays or filter the fireflies post-render, neither of which are ideal. Using more modern techniques such as VCM PTC can solve the issue, at the cost of render time, in general.

kolby
06-13-2018, 09:23 AM
Lot of the fireflies comes from very bright and small light sources. Try simulate distant light with the use of area light. It could help in some cases.

Kryslin
06-13-2018, 09:54 AM
You can also increase the angle of the distant light to 20 or so. Someone mentioned that this helped them with fireflies.

Tobian
06-13-2018, 10:17 AM
Well it would because that would decrease the intensity of the light by several orders of magnitude. It's not a good fix.

Mats Andersen
06-14-2018, 01:30 AM
Yes typically i only use brute force now, as while it's slow it's predictable, and less prone to issues, if slower. I struggle with interpolated too tbh. yes, theoretically brute force can result in noisy indirect specular, it's mostly the reflections of reflections, from very bright (necessarily) light sources, which causes fireflies, which is why i was establishing if you had used old-standard, as that can magnify and make fireflies so much worse!

There's a few paths to solving indirect specular, from using bi-directional path tracing, and VCM, which resolve causicts much better than MC, raytracing, to simply clamping indirect specular, which produces less realistic but normalised results, which are easier to clean. TBH if you look into it every single PBR based renderer which handles indirect specular (and indirect diffuse to an extent) suffers from fireflies, because you cannot multiple importance sample indirect, and either clamp indirect rays or filter the fireflies post-render, neither of which are ideal. Using more modern techniques such as VCM PTC can solve the issue, at the cost of render time, in general.

Thanks guys! How to I approach such a solution? Is this built in options in LW?

Mats

Tobian
06-14-2018, 02:08 AM
Sadly we don't yet have pathtracing in LW, or indirect clamping.

The only thing we have is ray type switches, to swap out materials to specific rays

https://docs.lightwave3d.com/display/LW2018/Example+-+Ray+Type+Switch

Mats Andersen
06-15-2018, 01:48 AM
Sadly we don't yet have pathtracing in LW, or indirect clamping.

The only thing we have is ray type switches, to swap out materials to specific rays

https://docs.lightwave3d.com/display/LW2018/Example+-+Ray+Type+Switch

Looks complex but thanks for all the help!

M

Tobian
06-15-2018, 02:11 AM
It's fiddly but once you get it, you'll be 'aha!' :)

Pavlov
06-15-2018, 11:45 AM
Hi, no time to read this all but:
- never use an enviro for interior, use portals instead and set samples high (>20).
- most of the issues you see are caused by materials having Glossy on AND Raytrace + backdrop, so check that every material in scene has Raytrace only in reflection.
- A wrong glass can cause a noise feast. Try removing glass and let's see if this improves. If yes, i'll try to explain why and because.

Best
Paolo

lardbros
06-15-2018, 04:43 PM
Good advice Paolo :)