PDA

View Full Version : FG Multibounce Weirdness

Chris-TC
02-08-2006, 09:38 PM
Hi everybody,

I ran a couple of final gathering tests today and discovered something pretty strange.
I did these tests in XSI and I'm very interested whether my findings can be reproduced using Maya. Also I'd be interested if anyone knows what causes this behaviour.

So, here is the simple scene setup:
http://www.digipiction.de/var/multi1.jpg

It is a basic corridor. The walls are spaced 3 meters apart, the ceiling is 3 meters high. The material for walls, ceiling and floor is a simple Lambert (50% grey) without reflectivity or anything fancy.
In the screenshot you can see where the camera is pointing. At the end of the corridor I have placed a block with a constant white material and a brightness of 200. This ensures that the block generates final gathering. There's no light in the scene.

So here we go, I'll post the relevant settings:
FG Bounce - 0
FG Reflection Trace Depth - 2
FG Refraction Trace Depth - 2
FG Total Trace Depth - 4
http://www.digipiction.de/var/multi2.jpg

At the default settings without multibounce turned on, all we see is black. This was to be expected.

If I keep increasing the bounce amount, I'll start to see a bit of illumination:
FG Bounce - 4
FG Reflection Trace Depth - 2
FG Refraction Trace Depth - 2
FG Total Trace Depth - 4
http://www.digipiction.de/var/multi3.jpg

What's interesting at this point is that if I add one more bounce, Mental Ray throws a warning:
warn 082069: finalgather diffuse depth 4 too high set to reflection 2 + refraction 2

I've ran several tests and found out that the amount of Reflection and Refraction Trace Depth needs to add up to the Total Trace Depth and the amount of multi bounces needs to be below this Total Trace Depth. Otherwise, Mental Ray will throw a warning.

Nevermind, let me not get sidetracked here. So you can see that at 4-2-2-4 we get a bit of illumination.
Now it gets interesting. Check this out:
FG Bounce - 4
FG Reflection Trace Depth - 3
FG Refraction Trace Depth - 2
FG Total Trace Depth - 5
http://www.digipiction.de/var/multi4.jpg

I increased the Reflection Trace Depth and suddenly we get a lot more illumination! Once again, the lambert corridor is not reflective!

We can push it further:
FG Bounce - 4
FG Reflection Trace Depth - 4
FG Refraction Trace Depth - 2
FG Total Trace Depth - 6
http://www.digipiction.de/var/multi5.jpg

We get even more illumination. Now what happens if we decrease the reflection depth? Since nothing in the scene is reflective, you'd think that it doesn't matter.
But it does:
FG Bounce - 4
FG Reflection Trace Depth - 1
FG Refraction Trace Depth - 2
FG Total Trace Depth - 3
http://www.digipiction.de/var/multi6.jpg

Suddenly we get less illumination than originally - in this case the image turns practically black again.

Can anybody explain what's causing this behavior? Why does the Reflection Trace Depth influence the amount of illumination?

During these tests I also found out that final gathering is not ideal in extreme cases of bounced light. By adding a sphere into the scene and cranking up the settings I can get some nice illumination but it's still splotchy and the render times are insane. Photon mapping certainly would be faster in a case like this.
http://www.digipiction.de/var/multi7.jpg

Cheers,
Chris
francescaluce
02-08-2006, 09:50 PM
where's the weirdness ? :)
fg bounces(above 1) ARE reflection rays.

try in xsi a ray_type node using a custom
port for reflection rays.. also the fg will get
that color even if as you said there's no
reflecting objs. not only, but it is true also
the inverse.. if you color correct your bounces
with a ctrlirradiance.. it will color correct also
your reflections !

same thing for refraction rays where involved
ray depth. I hope this situation will resolved soon
when optimizing the bug mr has with transp objs..
as it could really belong to this. it seems to me
that is just a first implementation of multibounces
and that it will refined in future versions.

but in maya ie, strictly it is not weird as the interface
also let the user feel a certain relation between bounces
and reflect/refract (infact in the standard interface there're
no parameters to let you put fgbounces directly but instead
just using reflect/refract rays).. but in xsi you have to know
this before not considering it as it looks.. weird.

having said that there're no reasond to use bounce>1
for sure in this scene, as you already pointed out, just
because we have photons to do that, so not that
great matter.


ciao
francesca
Chris-TC
02-08-2006, 11:12 PM
So what you are saying is that the FG reflection depth limits the total amount of FG bounces, right?
I could set my FG bounces to 10, but with an FG reflection depth of 2 all I'd get is 2 bounces.
Is that correct?
dagon1978
02-08-2006, 11:33 PM
So what you are saying is that the FG reflection depth limits the total amount of FG bounces, right?
I could set my FG bounces to 10, but with an FG reflection depth of 2 all I'd get is 2 bounces.
Is that correct?

correct ;)
Chris-TC
02-09-2006, 03:46 PM
Okay, thanks for clearing that up guys!
bgawboy
02-11-2006, 03:37 PM
Just to clear up any confusion, with respect to the FG trace depths...

The FG diffuse depth, or number of 'multibounces', is the number of diffuse interactions, whether they be reflection or refraction.
The FG reflection depth, number of reflections, is the number of diffuse, glossy, or specular reflections spawned by an FG ray.
The FG refraction depth, number of refractions, is the number of diffuse, glossy, or specular refractions spawned by an FG ray.

The FG trace depth total is just an overall limit for the total number of interactions spawned by an FG ray.

So, if you have only multiple diffuse reflections, as in the case of objects using only the default Lambert shader, then the trace depth will be the lesser of the reflection and the diffuse depth numbers.

In the Maya UI, there is only a checkbox for diffuse interactions (multibounce). When it is translated, it comes out as 100, which is usually much higher than any other of the numbers. This makes it effectively an enable for diffuse reflection, not just specular and glossy, and then you only have to pay attention to your FG reflection trace depth number.

For this scene, I agree that GI (photon map) is probably better. This is usually the case when the level of multibounce goes much above 3. However, I have had scenes in which I preferred the control of the 4 or 5 multibounce FG over the GI photons. In this scene, I think you need to crank up your photon trace depths pretty high, probably over 10.
bgawboy
02-11-2006, 03:52 PM
And with GI, you'd convert your box into an area light with photon light emission.

Actually, FG+GI might give the best results, but you'd still need to make a photon light emission.
Chris-TC
02-11-2006, 04:01 PM
And with GI, you'd convert your box into an area light with photon light emission.

Actually, FG+GI might give the best results, but you'd still need to make a photon light emission.

Yes, in this kind of scene FG+GI does give the best results. It's not a scene I was actually working on, I had set it up just to do the testing.

Out of curiosity I also activated photon mapping afterwards (using an area light the same size as my box) and got pretty nice results almost immediately.
The default amount of photon bounces in XSI is 20 and it worked well in this scene. The good thing about photons is that additional bounces don't slow down the calculation a big deal.

Now of course you'd have to use a whole lot of photons to get a clean looking result in this scene, which is why the addition of FG makes sense to even out illumination and get very accurate shadow detail.
CGTalk Moderation
02-11-2006, 04:01 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.