Cleanup noise in Elements: Reflection and Shadow - Also LOOONG render times

[cecofuli]

Hi svetlozar,
yes, I inverted the description on my gif. Anyway, thanks for your time and new setting. But in my opinion there is a little, but important difference.
See the highlight on the chrome shader for the chair.
There are some jagged edges. And, for example, in animation, you will see some artifact and flickering in this area. See how smooth is with my settings.
But, after a lot of test, IMO there are no solution for Sub-pixel OFF on the highlight: jagged edge and/or High rendertime.

[svetlozar.draganov]

Hi svetlozar,
yes, I inverted the description on my gif. Anyway, thanks for your time and new setting. But in my opinion there is a little, but important difference.
See the highlight on the chrome shader for the chair.
There are some jagged edge. And, for example, in animation, you will see some artifact and flickering in this area. See how smooth is with my setting.
But after a lot of test, there are no solution for Sub-pixel ON on the highlight: jagged edge and/or High rendertime.

Yes you are right - there is a jagged edge on the chair with Sub-Pixel OFF - we will try to tweak a little bit the shader for this object.

[Dariusz Makowski (Dadal)]

Howdy, I've been following this thread on/off for a while now, is this scene maybe available somewhere to download ? - I'd like to run some tests...

Thanks, ciao.

[cecofuli]

Search here =)

[joconnell]

Hi Svetlozar,

Just a few quick questions to see am I thinking clearly about sampling!

Okay so at the start vray shoots one AA ray, this hits a surface and then the surface starts doing material, sss and lighting sampling. The material samples are determined by how many subdivs you've set in the material, and likewise the light samples. Are all of these samples being tested by the noise threshold of the DMC sampler first to see if they're good enough, and then fed back into the AA sampler? Does the aa sampler then determine if the final pixels are good enough via it's noise threshold and based on that make a decision about making more samples? If it's not good enough, the AA sampler fires two rays and then halves the material and lighting subdivs between each. So in theory if you don't reach your noise threshold and the max number of samples are used, the second AA pass will use pretty much the exact same number of samples as the first AA pass and again barring early termination probably take a similar amount of time? Do the results of the aa 1 pass and the aa 2 pass get added together and averaged?

What I'm thinking is that the AA sampler can ultimately clean up any noise in the scene, but that's probably a bad way to do it. For example in the scene with the glossy materials that morne posted, the majority of the noise is coming from the materials so if you're using AA 1, 100 you're going to end up going through a lot of loops of AA sampling, whereas if you had given the renderer enough Material samples right at the start (forced by low aa) you'd get a clean result at a much earlier stage.

Does the slow down in renders come from Vray having to go through numerous loops of increasing AA samples more than anything else?

[zeronove]

Hi Svetlozar,
thanks for your answer, very informative, as usual ...let me ask you the last one about LWF then I'll try to stop me boring you:

talking 'bout first LWF scenario (no compositing), which is your suggested state for "Affect Color Selector" and "Affect Material Editor checkboxes? (I mean the ones inside the max preference setting dialog).

I'm going to test further this new way (for me) to approch lwf (using vray gamma instead of max gamma) but, let me say, that if it will turn to be a viable option, I'll suggest to chaos writing the specific lines in the help file in a different way because actually it seems to me lwf checkbox is a big NO with the exception of some circumstances.
That's what's written there: "Note that this option is intended to be used only for quickly converting old scenes which are not set up with proper linear workflow in mind. This option is not a replacement for proper linear workflow."
Maybe it's my language knowledge to be lacking but beacause of those lines I've never tried that way.

[joconnell]

Also as far as I know the Linear workflow doesn't affect non vray materials too and things like textures inside of vray light materials?

[trixian]

It does if you use the max gamma and material\swatch settings.....that is as far as non vray materials go. The map inside a vraylight material I'm not sure about, but I'd hope it behaved consistently like with other maps.

[svetlozar.draganov]

Hello guys - here is a link to a video where you can see why I prefer to work with Vray Gamma instead of 3DSMax gamma.

talking 'bout first LWF scenario (no compositing), which is your suggested state for "Affect Color Selector" and "Affect Material Editor checkboxes? (I mean the ones inside the max preference setting dialog).

That's what's written there: "Note that this option is intended to be used only for quickly converting old scenes which are not set up with proper linear workflow in mind. This option is not a replacement for proper linear workflow."

When you are using Vray Gamma setup and 3DSMax Gamma is of - "Affect Color Selector" and "Affect Material Editor" don't affect anything - they are only work when 3DSMax Gamma is ON.

About our help file - I described in one of my previous posts why is written this way.

Also as far as I know the Linear workflow doesn't affect non vray materials too and things like textures inside of vray light materials?

Vray Gamma Setup is working on Vray Materials - Standard Materials are not gamma corrected.
Vray Light Material texture is gamma corrected!

[Morne]

this is what I don't understand. Vlado all the time say NOT to use that "magic" LWF button as its not magic, but instead use affect "dont affect colours".
2 or 3 years ago when ChaosGroup was doing a demo in South Africa, the guy from Chaosgroup that was here also said to just use that button and ignore the max gamma.

Now I ask you, why the conflicting info? Also, if you use the "linear" magic button and NOT the "dont affect colours", wont this mess up your LWF exr and vrimg output so that when you open in another app, its all messed up? (I can test it myself, but my pc is busy right now)

[svetlozar.draganov]

Hi Svetlozar,

Just a few quick questions to see am I thinking clearly about sampling!

Okay so at the start vray shoots one AA ray, this hits a surface and then the surface starts doing material, sss and lighting sampling. The material samples are determined by how many subdivs you've set in the material, and likewise the light samples. Are all of these samples being tested by the noise threshold of the DMC sampler first to see if they're good enough, and then fed back into the AA sampler? Does the aa sampler then determine if the final pixels are good enough via it's noise threshold and based on that make a decision about making more samples? If it's not good enough, the AA sampler fires two rays and then halves the material and lighting subdivs between each. So in theory if you don't reach your noise threshold and the max number of samples are used, the second AA pass will use pretty much the exact same number of samples as the first AA pass and again barring early termination probably take a similar amount of time? Do the results of the aa 1 pass and the aa 2 pass get added together and averaged?

What I'm thinking is that the AA sampler can ultimately clean up any noise in the scene, but that's probably a bad way to do it. For example in the scene with the glossy materials that morne posted, the majority of the noise is coming from the materials so if you're using AA 1, 100 you're going to end up going through a lot of loops of AA sampling, whereas if you had given the renderer enough Material samples right at the start (forced by low aa) you'd get a clean result at a much earlier stage.

Does the slow down in renders come from Vray having to go through numerous loops of increasing AA samples more than anything else?

DMC Sampler works progressively - it doesn't test the max amount of samples added into material parameters at the beginning.
For example if you have material subdivision= 8 = 8*8 = 64 Samples ,
Let's say that DMC Sampler has this values - Adaptive Amount = 0.85 and Noise Threshold = 0.01 and Min Samples = 8

With Adaptive Amount we determine how much adaptive will be Vray - 0.85 means the Vray will be 85% adaptive - i.e. Vray will try to calculate 25% of the samples we've set in Material Subdivision. 25%*64 = 16 Samples - These are the samples at the beginning - of course if this number is lower than Min Samples Value - DMC will start from Min Samples Value.

Noise Threshold will determine if this 16 samples are good enough or not - if there are not, Vray will start shooting more samples , until Noise Threshold say that the samples are good enough or if Max Subdivision Value is reached.

This calculation will be calculated for every Camera Ray (AA ray)


Adaptive DMC Image Sampler is working similar way as DMC Sampler.

Let's say you have Min Subdivision= 1 , Max Subdivision = 10 , Color Threshold = 0.01
At the beginning Vray will shoot 1 Sample from each pixel - then Color Threshold will determine if this samples are enough or not - and if they are not, Vray will shoot more samples progressively from this pixel - the place for each sample in the pixel is not random - Vray will try to add samples in such way to achieve best results with less samples - after each shooting Vray merge the new samples with the previous ones and run the Color Threshold Check - and the procedure continue until Color Threshold decides that the samples are enough or the Max Samples are reached.

If the Color Threshold is very low as 0 - Vray won't shoot all the Samples during the "first pass" - Vray will always shoot samples on steps - If you have Min SubD = 1 - Max SubD=10 Vray will works this way:

1.First Pass = 1*1 Subdivision = 1 Sample
2.Second Pass = 2 Subdivisions = 2*2 = Samples = 3 New + 1 from previous pass
3.Third Pass = 3 Subdivisions = 3*3 = 9 Samples = 5 New + 4 from previous pass

Tenth Pass = 10*10 = 100 Samples = 19 New - 81 from previous pass

With the universal settings Min = 1 , Max = 100 and with Color Threshold to control the quality of the image - you can achieve good results in most cases - but the render time won't be optimized at all. The good thing about this workflow is that you don't have to adjust any subdivisions in your scene - everything is controlled by the Image Sampler - Lights , Materials , DOF , Motion Blur etc . In other words DMC Sampler is ignored when you are using Universal Settings! The final image will be good in most cases - but not as good as if you adjust every parameter individually . The render time will be significantly longer.

Let us know if you have any questions about the above information.
Don't hesitate to ask us if you need further support about our products.

Regards
Svetlozar Draganov

[cecofuli]

Svetlozar, you are right, but sometimes we don't have the time to try 30 different setting. To change the IM setting, change the subdivs for DOF, Glossy, MBlur, make tons test rendering and see if looks nice. We prefer to know that, if I click render, all is perfect and fine. I prefer use 3 hours on my machine instead 3 human hours to check the correct settings. This is our goals.

[Morbid Angel]

Svetlozar, you are right, but sometimes we don't have the time to try 30 different setting. To change the IM setting, change the subdivs for DOF, Glossy, MBlur, make tons test rendering and see if looks nice. We prefer to know that, if I click render, all is perfect and fine. I prefer use 3 hours on my machine instead 3 human hours to check the correct settings. This is our goals.

still there is no solution for your scene? I did some tests, and as well made an optimized render but also got the same badly pixelated highlites. Those damn ies!

[cecofuli]

No Morbid, no solution! At the end, I use US with SUB-PIXEL and CLAMP OFF with a lot of hours. 14 hours or more with one I7 esacore at 4.0 Ghz (2500 px, so not very high)
But you know, the buckets were crazy when they were above the chrome and glass shaders!!! Also THE usual annoying problem with jagged edge on ceiling lights. This is the RAW final image:



DOWNLOAD

And, also, I've one scene that it was IMPOSSIBLE (really, impossible) to render this image with US and sub-pixel and clamp OFF.
Then when buckets ( 8x8 ) are on the glass lustre (I model it by myself) are completely unmovable! See the next image.
The full image (original was 2500x1500px) with US and sub-pixel and clamp ON (Noise 0.007) with one esacore 4.0 Ghz in about 3 hours.
The little test with sub-pixel and clamp OFF (Noise 0.05) with 4 esacore 4.0 Ghz in about 15 minutes at 1024x640!!!!
And, again, the problem was on the little, bright, reflection pixels! Sometime only for one or two bright pixels, one single bucket ( 8x8x!!!) takes 3 minutes .
If you have the solution, please, write me =) PS: "Affect specular" on the IES is off and for the little VRayLight plane as a spotlight, Affect reflections" is OFF

[zeronove]

Svetlozar,

thanks a lot for your very informative and detailed answer, I really appreciate them.
Even if it's a fact every one of us is always secretly dreaming for the "make it good" button ...I'm still very interested in these kind of infos allowing me to deep edit and optimize my scenes
So I'll give for sure a serious try to the "forgotten" lwf checkbox, just to see if it works for me as well.