light cache and glossies

[vance3d]

I am wondering how the settings in lightcache effect the quality of rendered glossy reflections especially when computing a light cache for a flythrough animation.

in the material settings described in the manual it only mentions the renderer using the lightcache if the setting for treat glossy rays as gi rays is set to always.

so I'd like to know how many of you change the settings in the materials for glossy ray treatment and what (if anything) you need to take into account in the lightcache with respect to how its being used for glossy reflections.

thanks,

V Miller

[Clifton Santiago]

I recently finished an IR/LC exterior animation and I used LC for glossies. It worked great. I couldn't see any difference from calculating it normally. And because I put glossy reflections on almost every surface, it really sped up my frame time.

The only time I've noticed that LC and glossies don't work as well is with glossy refractions, like for a lampshade, or frosted glass. It seems to make them more absorptive and noisier.

[vance3d]

Clifton,

did you change your materials settings to the "always use gi rays for glossy rays" or did you just leave it at its default?

What kind of speed increases do you think you got out of using LC for glossy rays?

Finally, did you precompute your IR and LC? if you did, did you use the LC during the calc of the IR? And did you still keep LC in the secondary GI slot after computing the IR?

thanks for your help.

V Miller

[Clifton Santiago]

Yes, I enabled the "Use lightcache for glossy rays" option.

On one sequence, I found that my frame times went from 11-12 min to 6-7 minutes when switching to LC for glossies.

I precalc'd my IR and LC together, IR with multiframe incremental, and LC with flythrough mode. And I used both of them for the animation, loaded from the precalc files. You have to load the LC map if you are using it for glossies.

I precalc'd the IR every 10th frame and the LC is automatically calculated for every frame. Keep in mind the IR is geometry dependent, but the LC is time dependent so don't change your animation length after calculating LC.

[ene.xis]

i was under the inpression that if u set ur LC to flytrough u just need one frame to calculate it...right?

[Clifton Santiago]

It calculates the whole LC map for the animation, in the first frame. I just ran an IR/LC prepass yesterday, on one of my render nodes, via backburner, and the first "frame" took half an hour, and the remaining Nth frames only took 30 seconds each. Most of this time on the first frame was from the LC pass running for the entire length of the animation.

[dlparisi]

Clifton,
How did you go about selecting a sample subdiv's for the flythrough mode? I believe the example in the docs set it to around 3000 but I really don't know how this number was determined since I don't really understand what it does with those 3000 samples. Unlike a normal lightcache for a single image where the you can relate the samples to the final image, since it's computing it for the entire animation I can't get my head around how to determine the proper number of samples. Does it split the 3000 sample between each frame of the animation (e.g. a 300 frame aniamtion would get 100 sample per frame-sounds way to low) or is it somewhat adaptive like the Irr. map in that it'll start to fill in samples where there are gaps? Of course trial and error would work but since the flythrough calculations take so long this is a pretty slow process. Any hints or clues?

[Clifton Santiago]

I just use something between about 800 and 1200. Doing like 3200 like in Vlado's tutorial takes forever when you have a lot of lights, like in an interior. Occasionally I do tests with much higher LC subdivs, and its often difficult to see the difference in quality vs. the vast increase in calc time.

[dlparisi]

Doing like 3200 like in Vlado's tutorial takes forever when you have a lot of lights, like in an interior.

Exactly, that damn progress moves so slow when I try it at 3000 on a complex interior . I'll try it with a much lower setting and see how my results are. Thanks.

[flipside]

Do note that the 3000 samples are divided across the whole animation, while with a still they are only placed on the current view. So you might need more samples in a flytrough than for a still.

A good way to test LC is to set it for first and second bounce, then turn off any filtering for LC. You will clearly see the size and colors of each LC sample in the rendering and you can judge if it fits your needs. The size of the samples is set by the size parameter (clever no?), and the accuracy of the sample color is set by the subdivs. So to test the size, simply use a low subdiv value. For flythrough, use a scene dependant size and not screen or your sample size will vary each frame when you approach walls etc...

How many subdivs for your lc also depends on how many areas in your scene should be lit with second bounce light only. If almost everything is covered by the first bounce anyway (I mean if 90% of the brightness comes from 1st bounce), you can get away with low subdivs for 2nd since the effect on total lighting is so small and splotches won't be visible. But if you're in a room with small windows, and no artificial lights, you will need lots of subdivs to get the gi nice enough. However most of the arch scenes will usually have big windows, and if not, you would probably use indoor lighting to compensate for the low light situation.

If like clifton, you have glossies all over the place, the effect of use lc for glossies will be huge.

But what you say about the material option 'treat glossy ray as GI rays', I have no idea. I think most people misread your post thinking you were only talking about the general lightcache option 'use LC for glossies'.

So I would like to know the answer to that question too. Does it matter or not?

[Clifton Santiago]

Thanks for the tips flipside. What I have been doing for still previews is use the screen mode and before I go home, I switch to world mode for the animation mode, change the sample size, but not the subdivs. I guess I need to find the ideal subdivs specifically for my animation sequence in world mode.

[ene.xis]

nicely put flipside
thanx!
still i d like to have, and since everything is now so mathematical ( ), a rule thumb to have a less empiric way of finding the golden samples number for each animated solution...

[vance3d]

Flipside,

the material settings is exactly the issue I really wanted to find out about. The way they are described in the manual indicates that they may be used for different scenerios. I'd just like to have some clarification.

V Miller

[flipside]

I'm gonna do some test with it. (when I have some more spare time)