Vray GI and Color transmittance tests [maxwell] involved

[Gijs]

Right, then you found the right thread, because that's what it is about.
It is not about matching that photo, it is about how the render engines react.

The main reason I want to contribute to this thread, is because of the 'linear workflow' part of the story.

[Morbid Angel]

ok...now that we have that straghtned out lets focus on some interesting facts:

Ive got a simple setup of a box inside of which I got an emitter and a sphere. No GI is used at all. Point of this is to see how much diffusion or primary light is transmitted from the illumination source to the object.
I set the scene with appropriate scale i.e. the box is 3 m high etc. First set of images is default illumination in vray and maxwell with gamma at 1.0
We can see that both images are very dark, but in maxwells there is a bit more illumination. Second set is gamma at 2.2 and we can see that both images are now illuminated however again vrays is darker. Same goes for last set of images. One can say that vray light is weaker, however I can state that by the amount of luminance bleed on the wall the vray light is very close to maxwell.
Scene is here.
http://www3.telus.net/public/vinnik8...nce_test01.rar

[RErender]

It apears to me after a few renders that the vray light output is about 1/2 that of the maxwell light with a multiplier of 1000.

That being said, as I understand it, when normalize intensity is on, a
multiplier of 1 equals 1 watt per meter squared with system units set to one
meter. So with your file being set to cm as the system unit a multiplier of
1000 should equal .1 watt per meter squared or about a 1.2 watt bulb.

So what does Maxwell mean by a 1000 watt light?

Who is going to build a 3 meter room , paint it 50% gray and hang a bulb in it?

[vlado]

You have to consider what units your get in your final image too (e.g. what values do the pixels represent?). For V-Ray, the answer is easy - each pixel represents average radiance in W/(sr * m^2) - watts per steradian per meter squared, if no color mapping is applied. This is the same units that you specify the light intensity in when "Normalize intensity" is OFF. That means that a light with pure white color and multiplier of 0.5 (normalize intensity "off") will appear exactly medium gray on the screen if no color mapping is applied.

As for the "normalize intensity" option, I probably have to revise it to make sure it really applies the correct conversion when turned on.

Best regards,
Vlado

[dennis]

this is exactly the problem i stated in the other maxwell/vray comparison lights decay to fast and to avoid to dark areas, lights has to be multiplied more and will introduce burns in light entrance areas.

[vlado]

Here is an example:

Rendered image:


Light settings (color is 255, 255, 255):


Material settings for the base plane (diffuse color is 255, 255, 255):


This is the correct result as far as the Rendering Equation is concerned.

Best regards,
Vlado

[vlado]

this is exactly the problem i stated in the other maxwell/vray comparison lights decay to fast and to avoid to dark areas, lights has to be multiplied more and will introduce burns in light entrance areas.

The light decay is quite correct though in V-Ray.

Best regards,
Vlado

[Gijs]

'quite' correct Vlado? I assume that's an understatement?

Anyhow, dennis, I think light decay in Vray itself is not false, but the displaying of it, hence the introduction of gamma correction

[vlado]

Here is a test of light decay.

The light settings are below; the light is quite small and can be considered as very close to a point light. According to the Render Equation, the illumination on surfaces facing the light should be quite close to L/(d^2) where L is the light intensity and d is the distance to the surface. In our case, L is 1250.


In this first image, the plane has a pure white material and is 50 units down from the light. From the formula above, the amount of light falling of the plane directly below the source and reflected on the image should be 1250/(50*50)=1250/2500=0.5 or RGB 128, 128, 128 - which it is in the image below:


In this second image, the plane is 100 units away from the light, which means that the point directly below the source should be 1250/(100*100)=0.125 or RGB 32, 32, 32 - which it really is in the rendering below:


Best regards,
Vlado

[vlado]

This is a test of light bouncing.

The scene is a small spherical light surrounded by a sphere with a 50% grey material and a radius of 200.0 units, and normals are inverted.

The spherical light source is exactly in the center of the sphere, has a radius of 1 unit and surface radiance of 20000.0 W/(sr * m^2).

Here is what the scene looks like:



The light settings:


The sphere material:


Since the source is in the center of the sphere, the light always falls perpendicular to the surface, and we can use the formula from the previous example to compute the light arriving on the surface of the sphere: 20000/(200*200)=20000/40000=0.5. Since the sphere has a 50% grey material, this needs to be multiplied by 0.5, and we get the radiance reflected from the sphere on the image - 0.25 or RGB 64, 64, 64 which is exactly what we get in the rendering:


Now, we turn on GI. I used the light cache as a secondary engine and the irradiance map as a primary engine, since this gives us unlimited amount of light bounces. As light bounces around the scene, each time it hits the sphere surface, it is reduced by half (since the material is 50% grey). So, after the first bounce, we have 0.25+0.25/2, after the second bounce, we get 0.25+0.25/2+0.25/4, and so on. The actual calculation involves the rendering equation and is a little more complicated, but this is the end result. So, we have an infinite sum of numbers, but you can see that they form a geometric progression with a ratio of 0.5. In this specific case, there is a formula that allows us to compute that infinite sum, and this formula gives 0.5 for the final result. 0.5 corresponds to RGB 128, 128, 128 and is exactly what we get in the final rendering:


Best regards,
Vlado

[vlado]

I can use these tests to verify that V-Ray computes illumination correctly, since I know both the input data units (surface radiance for the lights) and the output units (average radiance inside the pixels of the image) and I know that there is a 1:1 correspondence between the image and the actual physical units.

If any sort of color mapping is applied, be it gamma correction, exponential mapping etc, this is no longer true.

These tests would be somewhat difficult to analyse in Maxwell, since you don't know what you get in the final image. Since Maxwell tries to model a physical camera, many other things come into play - e.g. vignetting, shutter speed, f-stop, gamma correction, film sensitivity (ISO), lens material etc.

Best regards,
Vlado

[jujubee]

How much does atmospherics factor in (if at all) for both renders?

[vlado]

How much does atmospherics factor in (if at all) for both renders?

Which atmospherics?

Best regards,
Vlado

[Morbid Angel]

Vlado, glad you could join us in this thread. Very valuable info. I quickly read through what you written and I understand that to get the correct illumination from vray lights you need a precise set of units in meters. So, If I have a scene set in cm, the value should be 0.1 instead of 1.0 ?
The reason for this, is when in max 1 unit = 1 m, the scene scale becomes very small and is impossible to work with. So that is why I use this setup.

Rerender yes I know what you mean, but, we'r doing testing here nothing more.

[dlparisi]

As for the "normalize intensity" option, I probably have to revise it to make sure it really applies the correct conversion when turned on.

Thank you, thank you, thank you!! You probably saw all of the confusion this caused in another thread, so having vray "normalize" the value correctly behind the scenes would be a great change.

David