does size really matter ?

[bernhard]

Hello friends,

I know I know... this funny title will be answered with yes... (especially from all our wonderful ladies out there)... allright allright... lol:

but guys... I am asking that question regarding V-Ray's render calculation.

What do I mean ?
Let's say, we create a scene file, where we build 10 houses.
The first one, is as tiny as an ant - the middle one, number 5, is as big as it should be.
And house number 10, is a planet.

I am just curios.. if we would have a file like that.. and the SIZE is so crazy mixed..
would it be render fine, esepcailly for animation ????

I was thinking how Vray could handle it.. especially with GI, and the light values and how it would influence an animation process for rendering.... flickering or any other issues for animation.


So what are your thoughts about that now:
Does size really matter ?


thx a lot for all your (vray) minds,
bernhard

[Morne]

If you're using them in the same scene with sun, sky, physcam my geuss is that it would matter. (I'm just geussing) Since people keep on telling they are suppose to be physically accurate. That makes me think yes size does matter.

Anybody know for sure?

[instinct]

It already matters before rendering due to precision problems.

Floating point values do not have a fixed precision (hence the name :P) wich basically means the larger a value gets the less precision you have.

Wich means if you have fast animation VERY far from the origin you will get some damn nice stuttering and other issues.

So yes size does matter...that's also the reason for the VRayError "Scene bounding box too large" in the logs if i am not mistaken. Huge values = bad for precision

Regards,
Thorsten

[rivoli]

as Thorsten says. which is why, as an example, it's always safe when importing any drawing from autocad to move everything close to the orign. just to avoid problems later on when rendering.

[^Lele^]

VRay is a raytracer (duh me!), so imagine what a small object, very very far from the camera, would do to a ray trying to intersect it.
In the space of one pixel, you may have meters, if not kilometers worth of geometry.
The poor Vray would have an extremely hard time trying to figure out with which color to represent that one pixel, and if the camera just moves a small fraction, the resulting pixel color may vary wildly (it may then not intersect anything at all, or intersect another object entirely).
This generally results in flickering and related issues.
Hence the VRay warning.

[Sawyer]

So Lele then are the rays cast from 0,0 radiating out? What you posted makes sense but I am not sure how it relates to the origin warning.

[^Lele^]

VRay is a raytracer (duh me!), so imagine what a small object, very very far from the camera, would do to a ray trying to intersect it.
...
Hence the VRay warning.

The vray warning i was referring to was the one talking about scene bounding box.
That may have your object dead in the center of the world, and your camera a zillion units away, which would result in a huge scene BBox.
Ie. at least one object and your point of view are very far from each other.

The Origin issue has to do with max not being a full FP app, internally (don't quote me on it, but i believe it's a s 16bit half-float), and hence introducing errors when you reach the (very sharply coming, these days) limits of what 16bit math can offer you.
Even with VRay, which is 32Bit (i believe. Vlado?), if you do do math at the very limit of 32bit precision, the approximation errors tend to pile up, and there's little Vray (or anything else that works with 32bit precision) can do.

[Sawyer]

Thanks Lele. Makes sense.

[bernhard]

hmm...

Hello my friends,

thx a lot for your posts and thoughts, I really appreciate that.

Well.. one ting is for sure... the Unit setup always matters... in my opinion especailly for dynamics like reactor, fluids, cloth, forces, fire, smoke, explosions or of course for rigging and skinning and so on.

But I am not totally sure if the actual geometry size (independet from the Unit setup) causes render issues like flickering. Actually, I suppose .... it shouldn't.

Of course you would need to adjust light values, dependet to the size.
So.. let's say you would have a house in a unit setup that matches real life measurement.
And you create a VrayLight with value 1. Light source is okay and simulates the power of an interior light.

And now you shrink the house to the size of an ant... and you would leave the same light source and value it will be extreme bright. You would need to make the light source smaller (with the real width/lenght factor so max recognize it) - so you would match again the real life simulation.


But here it comes.. if you would create such an extreme example... and let's say you go from a microscope small piece to the size of a planet... how would vray deal with that ?

hmm.. unfortunately I don't have time right now to make some tests in a practical way, but the topic alone makes me curios....


thx for all you gret comments,
great people - great forum - like always,

happy rendering,
bernhard

[bardo]

Even with VRay, which is 32Bit (i believe. Vlado?),

I'm not actually sure but some time ago I've read that Vlado said that Vray uses 64bit math estimations...

[instinct]

As said this will cause issues. Mind you that max doesnt use the unit setup internatlly.
You can check out the problem when going to unit setup -> System Unit setup.
You have a slider to visualise the prblem. Just drag it to the right. And you get the values. When away 16777215,0 from the origin (not the camer, the world origin) you are faced with an acuracy of 1 Unit. Basically meaning the camera will move in 1 unit steps. This is for animation and object placement. But the same is valid for rays of course. This also creates all kinds of weirdnesss when using moblur for example...been there done it and slapped my head on the table :P

Regards,
Thorsten

[MasterBercon]

Vray uses 32bit accuracy with rays, max uses the same. Using 64bit floatting point values would give more accuracy but if the scene is correctly scaled ie. no huge or too small scale it doesnt really matter which you use. Using 64bit values would nearly double the required memory for meshes etc. so...

[bernhard]

hm...

well., well.. appreciate all your comments...,

till later,
bernhard