Forum Groups
  All forums
    Help & Feedback
      Questions
      Work in progress
      Finished Art
      Non-Max related
    Community
      Offtopic
      News
    Hardware
    Photography



Maxunderground news unavailable

 
First page  Go to the previous page   [01]  [02]  Go to the next page  Last page
 
A better understanding of Vray DMC
show user profile  3joez
I'm trying to master all informations about DMC sampler in Vray. But I think its logic is quite twisted.
So here I'm asking some questions to all you, Vray wizards.

1) Dmc sampler can be detached from antialias? I mean, they are in the same tab and often
people tend to describe the two things as related.
Shouldn't the sampler only sample the color and light information related to the raytraced ray,
starting from the camera and hitting the object?
And should the antialias filter be applied after that process, if I tick, say area filter?

2) In this famous article, already flying in other posts,
http://www.interstation3d.com/tutorials/vray_dmc_sampler/demistyfing_dmc.html

is said that if you have 50 glossy subdiv on a material and put 10 in the maximum subdiv of dmc,
you get 5 subdivision. So, lower quality, since it doesn't oversample the original 50 value, instead,
it seems that it undersamples it.

Following this logic, the less I subdivide, the closer I'm getting to the original 50 subdivision.
If I put 1 in max DMC I get 50 glossy subdiv, and probably a beautiful, clean, reflection.
If I put 20, I have 2,5 subdivs and probably an ugly, dirt reflections.

BUT, when I increase maximum subdivisions, I see that quality gets better, so it cannot be as it is saying.

Or, like probably is, my deduction must fail at some point.
I know I've put it quite complicated but I know you guys are understanding.
I will be very happy to hear opinions, on that.
Photobucket

read 3364 times
12/15/2011 2:59:28 PM (last edit: 12/15/2011 2:59:28 PM)
show user profile  Garp
You get 5 subdivisions per ray.
As I see it, min/max DMC sampler values deal with geometry (rays sent from the camera) while local sampling values set the limit of geometric rays times subdivisions (rays sent from the surface, whether for materials, shadows, DOF, etc).
With all this influenced by the amount of adaptiveness.
So depending on the scene - flat surfaces with detailed textures or tiny geometric details - you'll go for a low max value and adjust locally or go for something like the universal settings. Tricky stuff.




read 3330 times
12/15/2011 10:38:46 PM (last edit: 12/15/2011 10:38:46 PM)
show user profile  3joez
So, dmc controls global subdivisions, while materials, lights, dof, etc, CAN be controlled locally?
This come in handy when you have a lot of simple materials and say, just one with glossy reflections. Is it right?
That way you can subdivide the glossy two times, one with material subdivision and the other, with dmc?
Photobucket

read 3307 times
12/19/2011 10:49:37 AM (last edit: 12/19/2011 10:49:37 AM)
show user profile  Garp
min/max values for the DMC sampler control how many rays are cast from the camera to a part of the scene that is covered by one pixel.
Let's say that that particular area has a material with a blurred reflection applied to it.
The reflection subdivs value of that material controls how many samples are shot from this area (corresponding to one pixel) to the surroundings.
So you could say that the primary rays (controlled by the DMC sampler) sample the geometry and texture they hit while the secondary rays (controlled by the material) sample what is to be reflected in the surface hit by the primary rays.

From what I understand, the material's subdivs value defines the number of samples (rays shot from the surface) no matter how many primary rays (from the camera) hit that area.
So with subdivs = 8 - that is, 64 samples are to be taken -,
- if 1 ray is shot from the camera, that ray is then split in 64 rays to sample the surroundings;
- if 4 rays are shot from the camera, each one is split into 16, so there are still 4 x 16 = 64 samples taken from that area;
- if 64 rays are shot, then each one results in 1 secondary ray. Again 64 samples.

The problem is that the quality of the blurred reflection seems to differ for the same subdivs value.
I've experimented a little but it's kind of hard to pinpoint. I'm still trying to make sense of it so take my conclusions with a pinch of salt.
Still using the same area corresponding to one particular pixel, let's say 1 single ray is cast from the camera and is then split into 64. The secundary rays are spread into a cone whose angle is controlled by the glossiness value.
If 4 rays are coming from the camera, each one is divided into 16 rays and for each of the 4 groups, the 16 rays are again spread into a cone covering the same angle.
If now 64 rays come from the camera, obviously the unique secondary ray for each one of them cannot cover a cone and its direction has to be randomized. Otherwise it would be as if the glossiness were set to 1.0. The total number of samples for that pixel is still 64 but the result doesn't look as good as with the 64 samples shot from one incoming ray. I suppose only Vlado could tell what is exactly happening there.
My guess is that 64 samples taken from 1 primary ray will sample the reflection pretty evently. But 64 samples, each taken from one of 64 primary rays, will sample the reflection at random and the result will be noisier.
Hence the counter-intuitive result of getting a lower quality for the blurred reflection with more primary rays and the need to crank up the material's subdivs value a little to compensate.

But as I said, don't take my word for it. It's all guesswork based on some readings and experimentations. I might be completely off.




read 3301 times
12/19/2011 12:06:22 PM (last edit: 12/19/2011 12:08:40 PM)
show user profile  3joez
OK, Garp, soon I will post the same question to the chaosgroup forum...so I will tell you and meet there.
Meanwhile, what do you mean by "So with subdivs = 8 - that is, 64 samples are to be taken"?
What number is being subdivided, and what does it represent?
Photobucket

read 3284 times
12/19/2011 4:48:55 PM (last edit: 12/19/2011 4:48:55 PM)
show user profile  Garp
For reflection, the amount of samples taken is the square of the subdivs value (n subdivs -> n rays in U x n rays in V).
At least that's what I understand from this:


I don't know about shadow subdivs or other blurry effects but for consistency I suppose it's also the square of the value. Again, not sure.




read 3266 times
12/19/2011 11:17:06 PM (last edit: 12/19/2011 11:29:12 PM)
show user profile  3joez
Sometimes I think I'm dealing with all the parameters correctly, and then something unexpected pops out. Thanks to all answers, so far...

Garp you said that maybe with DMC you decide how many rays I shoot from the camera.
But Vray says "max and min SUBDIVISION", not "rays" (we are always dealing with raytracing, just to point it out). So what do you exactly subdivide?
Maybe the rays sent from the primary and secondary algorythms?

Photobucket

read 3250 times
12/20/2011 3:45:29 PM (last edit: 12/20/2011 3:45:29 PM)
show user profile  3joez
Well I'm up to something, I think.
I'm reading Legrenzi's Book and SUBDIVISIONS means that your vector inside 3dsmax is divided
a number of times to produce a raster image. That process is called SAMPLING.
As I see it, it's like "rasterize" in Photoshop, with the difference that in Vray you can control the quality
of the process.

Vray generally oversamples (subdivide more) when it encounters a lot of details and undersamples (subdivide less) when it encounters flat areas.

And more...sampler goes with antialiasing because the rasterization process is also called aliasing.
Alias means "other" but also "pseudonym", in latin, so aliasing is the technique to change a thing into a similar instance (vector into raster).

I'm just writing to consolidate my thoughts...tell me if I'm wrong.
Photobucket

read 3243 times
12/20/2011 4:41:46 PM (last edit: 12/20/2011 5:11:23 PM)
show user profile  Garp
>Garp you said that maybe with DMC you decide how many rays I shoot from the camera.
But Vray says "max and min SUBDIVISION", not "rays"

In the DMC sampler, it's the pixels that you subdivide. Say Min = Max = 2, then each pixel is subdivided by 2 x 2, so 4 rays are shot from the camera per pixel, each one allowing the computation of a sample. With Min = Max = 10, it's sampling the area corresponding to each pixel 100 (10 x 10) times by shooting 100 rays from the camera.


>Vray generally oversamples (subdivide more) when it encounters a lot of details and undersamples (subdivide less) when it encounters flat areas.

That depends on the sampler you're using. Adaptive subdivision can over- and under-sample. DMC can only over-sample. Both samplers will adjust the amount of samples according to the color threshold. So it's not about the area being flat or not, the color variation can come from the texture, what's reflected in the surface, differences in lighting, etc.


>Alias means "other" but also "pseudonym", in latin, so aliasing is the technique to change a thing into a similar instance (vector into raster).

Not exactly. From wikipedia:
"If the resolution is too low, the reconstructed image will differ from the original image, and an alias is seen. (...) Techniques that avoid such poor pixelizations are called anti-aliasing."




read 3237 times
12/20/2011 5:38:02 PM (last edit: 12/20/2011 5:38:02 PM)
show user profile  3joez
Everything is starting to make sense.
The key concept here is that DMC subdivides everything.
But its relation to scene values, isn't still clear, to me.

Let's imagine a scene with:
1)a simple material that has 8 reflection subdivision
2) a light that have 16 subdivision

dmc is set to min 2 and max 4

What values do you get?
Photobucket

read 3222 times
12/21/2011 10:19:04 AM (last edit: 12/21/2011 10:19:04 AM)
show user profile  Garp
DMC sampler: at most, 16 rays (4 x 4) per pixel are cast from the camera.
Mtl reflection: at most, 4 secondary reflection rays are cast per primary ray, making a total of 4 x 16 = 64 samples = 8 x 8 (mtl subdivs).
Area shadows: at most, 16 secondary light rays are cast per primary ray, making a total of 16 x 16 = 256 samples = 16 x 16 (light subdivs).

Let me say this again: this is my understanding and it could be wrong in many places; also it's based on sources that can have their on mistakes in them (the tut you linked to, Legrenzi's guide, etc). Don't take my word for it!




read 3217 times
12/21/2011 11:04:48 AM (last edit: 12/21/2011 11:12:31 AM)
show user profile  3joez
Seems I'm not alone in asking these things.
http://forums.cgsociety.org/archive/index.php/t-960256.html

> 4 secondary reflection rays are cast per primary ray

How do you say that?

Article says that material subdivision is divided by max dmc subdiv, so 8/4 leads to 2.
That 2 subdivision, add up to the original 8? Or does that mean the we have two subdivisions of 8 (64)?

I'm making images to understand better, seeing charts must clear things
Photobucket

read 3207 times
12/21/2011 3:47:57 PM (last edit: 12/21/2011 3:47:57 PM)
show user profile  Bolteon
sorry, but who fucking cares?


the write up on demystifying dmc is more than enough for information to get you understanding the system.


if you really want to know the whole thing; why not go study light/optical physics... and computer science. nothing person, but i'm baffled by what you'll gain by digging so deep.

-Marko Mandaric



read 3192 times
12/21/2011 9:18:12 PM (last edit: 12/21/2011 9:18:12 PM)
show user profile  Garp
> 4 secondary reflection rays are cast per primary ray

Read the rest of the sentence!
Ok, let's do it backward.
8 reflection subdivs means that a maximum of 64 samples (8 x 8) are to be taken for an area corresponding to one pixel.
4 max DMC subdivs means that a maximum of 16 rays (4 x 4) are cast from the camera for the same area.
In order to get the total of 64 samples, vray will cast at most 64 / 16 = 4 secondary rays per primary ray.

I'm starting to feel like I'm typing the same thing over and over.
And I kind of agree with Bolt. This is not something that will help you when setting up your render.
The DMC sampler settings control the quality of what's in front of the camera, the various local subdivs settings control the quality of their corresponding effect. The rest is purely academic.
You should really ask Vlado for this stuff.




read 3185 times
12/22/2011 2:28:46 AM (last edit: 12/22/2011 2:30:50 AM)
show user profile  Bolteon
*me high fives Garp*



"The DMC sampler settings control the quality of what's in front of the camera, the various local subdivs settings control the quality of their corresponding effect. The rest is purely academic."


perfect layman's term explanation. well done sir.

-Marko Mandaric



read 3177 times
12/22/2011 3:03:42 AM (last edit: 12/22/2011 3:03:42 AM)
 
First page  Go to the previous page   [01]  [02]  Go to the next page  Last page
 
#Maxforums IRC
Open chat window


Support Maxforums.org