Getting spanked w/C4D's GI: Help plz... 2 of 2

[kiwi]

I was just coming to tell you about MVs great radisoisty tute :shrug: .....hmmm I have no idea what is happening Paul.


Big stabs in the dark here,try messing around with the percentage for your smoothing tags and the diffuse fall off values for your materials.......big stab in the dark there though.



Arndts book has quite a bit on radiosity settings as well.



Sorry


Stu.

[Per-Anders]

up the accuracy to around or above 80%, the min and max samples are difficult, if you have lower values then it smoother out further between them and you can sometimes get smoother results, but on the other hand you need nice high levels otherwise you're going to end up with no detail and the spots/rings/artifacts seem to be higher contrast when using this solution.

alternatively use stochastic mode and bring teh samples up for a smoother result, but it's going to take an age to get right.

[AdamT]

A couple of things. First, you're not the only Cinema user at CGArchitect. Strat, who's pretty active over there, is a Cinema guy. Second, you need to be more methodical in your approach to radiosity. Looking at your settings, you're all over the place. Pick an accuracy (70 is *generally* good, but you need higher in a high-detail scene) and leave it alone. Why? Because accuracy is the sum of the Min and Max, so if you are changing all three numbers between renders it'll take a *long* time to get the optimal balance.

It looks like your scene has a lot of indirect lighting, meaning that you won't get very good light distribution with only 2 diffuse bounces. I'd up it to at least 5. If you have any light-emitting materials you'll need much higher samples. I prefer to use area lights rather than emissive materials in most cases.

Here's a general approach someone posted on Postforum (can't remember who--sorry):

"You have to start with low values (i always start with 20 stochastic samples, 1 min res and 10 max res) and remember that min res must be at least 1/10 of the max res (for example 3-30). I keep always high the accuracy value (between 85% and 95%), then I increase min and max resolution due to the distribution of stochastic samples and their density. When I have reached a good distribution I increase the number of stochastic samples et voilà. (this is not my personal interpretation, this is the explaining of finalRender official video tutorial)"

And here's some tips from Simon Wicker, who used to work for ILM and did some of the matte paintings in the Star Wars films, among others:

"Author: simonw ( ---.cvx40-bradley.dialup.earthlink.net )
Date: 07-21-02 12:06

hi kevin,

some further information to help you out. the shading artifacts in the xl implementation of radiosity tend to fall into two categories:

1) not enough shading points. if you have lots of very fine modelled geometric detail in you scene then you have to set your min/max values high enough to catch all of the detail. you can calculate this for yourself.

if your scene is 1000 units in size and you have fine detail 10 units in size then you have to set your max value to at least 100 (1000/100=10). you can use this calculation to determine some upper limits for your min/max values.

you can look for these errors in the corners of rooms or the fluting on columns.

2) not enough stochastic samples. these errors tend to appear as splotchy patches on flat surfaces. the radiosity engine calculates the illumination for each of the shading points generated and then interpolates the values to fill in the gaps.

however the quality of the shading point is determined by the number of stochastic samples that it sends out into the scene.the higher the samples the more likely it becomes that the illumination data for each point is accurate.

you can see this effect yourself. create a cube shaped room and stick in a spotlight to represent a light in the ceiling. set your render settings to 70% accuracy, 100 samples, 15/50, 3 bounces. now make some test renders where you increase the number of samples (200, 400, 800, 1600). you should see that as you increase the values you should get a much smoother result.

obviously how important this becomes depends on the underlying surface. a smooth white painted wall will need good sampling, while a nasty dirty grungy wall will probably not show any problems as everything gets hidden in the noise of the texture.

another determining factor is how complex the scene lighting is for the radiosity engine to solve. a room with a light shining in from the outside is considered complex because all the lighting is coming from an object out of the line of sight of the camera. in this situation all the lighting data has to be gathered by the stochastic samples sent out by the shading points.

our 'spotlight shining down in a cube room' is also a complex lighting situation as the shading points have to send out enough samples that they encounter the small patch of light falling on the floor many times to accurately sample the illumination available (90% of the surface area in the room is dark and only 10% is lit in any way but this 10% actually provides 100% of the illumination so the lighting is quite skewed).

cheers, simon w. "

[FusionDG]

Thx Stu for giving some thoughts on this…
Originally posted by kiwi
Arndts book has quite a bit on radiosity settings as well.


I’ve been looking at getting this book, I think Adam has said good things about it also. I was hoping a R8 would come out for C4D soon :shrug:

Originally posted by mdme_sadie
up the accuracy to around or above 80%, the min and max samples are difficult, … lower values …you can sometimes get smoother results, … high levels otherwise … no detail and the spots/rings/artifacts seem to be higher contrast when using this solution..

I have been leery of going that high with the accuracy, but both you and Adam feel that I should be around here, so I should stick with this and give it a try adjusting the other levels., thx

Originally posted by AdamT
A couple of things. First, you're not the only Cinema user at CGArchitect. Strat, who's pretty active over there, is a Cinema guy.

Strat’s been a bit quiet the last month or two of late. He’s been crit’ing some folks stuff this week, his work is actually a large contributor on why I chose C4D :D. He's been very helpful in the past to me.

Originally posted by AdamT
Looking at your settings, you're all over the place. Pick an accuracy (70 is *generally* good, but you need higher in a high-detail scene) and leave it alone. Why? Because accuracy is the sum of the Min and Max, so if you are changing all three numbers between renders it'll take a *long* time to get the optimal balance.


Yes, I guess looking back I was too eager to think I figured it out and started tweaking too many variables. The remaining info you posted is very helpful also. I’ll give these suggestions a try.

It seems that I have to start with a high accuracy setting, start low w/the min and max at a 1/10 ratio & work my way up, and make sure I have high quality Stoch count.

Thanks to all of you for your quick reply’s to my question.

Sincerely,

Paul

[AdamT]

I guess the really hard thing about radiosity is that it seems like every scene is different. Sometimes settings that render relatively quickly in one scene will take a decade in another scene. That's why you really have to be methodical in your approach. Just wish I could take my own advice. :)

BTW, I do highly recommend Arndt's book. The radiosity section is excellent.

[Mike Hooma]

your prepass should be at 1/1 (says so in the manual)

my experience has also been really frustrating with radiosity one render was taking 27 hours and it was only a small room still at 800*600 med radiosity settings, low accuracy tags on all small objects... I wish there was some better way, than just experimenting. I end up tidying it up in photoshop and living with the flicker for animations...

[MJV]

I see you're playing with the accuracy setting, which is pointless. The number of total min and max samples is a product of accuracy (which uses a curved exponential) times sample number. Accuracy has no effect on stachastic samples. At all. Period. So if you increase accuracy, all you're really doing is increasing the min and max samples value. Better to leave the accuracy setting alone and adjust those values manually so you know what's really going on. Use accuracy only to adjust min and max sample values for individual objects through the compositing tag. hth.

MV

[FusionDG]

Hi MV,

Nice to hear from you! You did an excellent job on your tutorial for radiosity in C4D. It explained a lot of things that I didn't glean from the C4D manuals.

One being that the accuracy only effects the min and max settings, and not the quality of the samples as the Stoch setting does. I think I got that right now, and a big part of that was your excellent web page examples.

I think Adam nailed my problem [one of them anyways :D] when he pointed out that I was all over the board in changing my variables. Looking back, I see what a waste of time that was!

I am running tests now and have gotten considerably improved results by using a 1/10 ratio on the min and max, setting the accuracy at one setting and leaving it [it's at 80 I believe] and I replaced my indirect lighting that I was relying on for my GI w/area lights at the windows. It seems to be giving me more "predictable light" if that’s possible?

The only thing that’s taking time is now I’m using a few lights w/area shadows. That’s a real time killer, nut it’s running over night and I’m saving the lighting solution, so hopefully the lighting solution will be useful for my other renderings on this house.

Thanks for everyone’s help, I’ll post my final results.

Sincerely,

Paul

[MJV]

I wouldn't use an area light in a GI scene, unless the render is for your grandchildren. Better to use a luminant object having the same area. Of course this will require that you increase stochastic samples considerably, the smaller the area the higher you will have to go, possibly in the 6000 range, but should still produce as good but faster results than using and area light. Perhaps in some circumstances an area light will be more time cost effective, but I would doubt it.

[FusionDG]

Originally posted by MJV
I wouldn't use an area light in a GI scene, unless the render is for your grandchildren. If you wouldn’t use AN area light in a GI scene, what about SEVEN AREA lights w/ five of them casting area shadows…:rolleyes:

When I came into work this morning, it looks like my estimated overnight render will turn into an over weekend render, if I am lucky, because my 1st render is about half way through the pre-pass w/19 hours already spent on the render, wow. And it doesn’t even look that much better. :annoyed:
Originally posted by MJV
Better to use a luminant object having the same area. Of course this will require that you increase stochastic samples considerably, the smaller the area the higher you will have to go, possibly in the 6000 range, but should still produce as good but faster results than using and area light. Perhaps in some circumstances an area light will be more time cost effective, but I would doubt it. Putting the Stoch in the 6000 range for an indirect scene I guess I can understand, because before it seemed that my indirect lighting method wasn’t producing the quality GI samples even at 1,500. And then once I started to turn up the min & max samples, it then got incrementally splotchy.

At this point I am thinking of putting some render tags on the light fixtures in the house [cans, lamps, etc] to omit them from the GI calculation and just run indirect lighting for teh GI at higher stoch samples or the area lights w/out the area shadows. I was getting pretty decent times and results before I added my light fixtures back into the mix and turned on the area shadows.

Thanks MV.

[Per-Anders]

woah, i didn't see area lights in there, yes, a big no-no, instead make some planes up, put a white material on them, and set their brightness in teh luminance channel way up (like a few thousand percent), that should help, then put a huge sphere around the scene and put in a grey blue luminance on that, but at a more normal setting... that hsould help, probably give it a gradient if you want the light bouncing from teh ground look (you know brown at the bottom of teh sphere, and blue at the top)... this will render a lot faster than using area lights, and still give you accurate shadows. don't forget to turn off autolight, i sometimes do things this way in radiosity mode raather than actually use area lights in a scene because they take so long to render.

[Darlock]

for me the min max ratio 140:150 makes things look better, and kills ugly shadows. and adding a couple of 0's is the last resort, but I never made any indoor scenes, so i cant say

[davedavidson]

ok here goes
my personal settings for interiors are this
acc: 90
diffuse 2
prepass: 1/1
samples 300
min 60
max80

i find this works well with my renders

[FusionDG]

Originally posted by mdme_sadie
...then put a huge sphere around the scene Hum, a sphere instead of a Sky Object [*EDIT*Skydome]? I think you mentioned this before somewhere [maybe here]. Is this because the sphere is more efficient than a Sky Object [*EDIT*Skydome]? If that is the case, then would it be even better to chop off the lower half of the sphere?
Originally posted by Darlock
for me the min max ratio 140:150 makes things look better, and kills ugly shadows. Well, I am getting better results with the 1/10 ratio. If things are still ugly I may try this
Originally posted by _dd
...my personal settings for interiors are this
acc: 90diffuse 2prepass: 1/1samples 300min 60max80 Can I ask what you use for a light source? I saw your pic "room" on your web page which looks like it is lit well.

Thx,

Paul

[davedavidson]

i always use a light where there is a light, treating it like real world
it seems to work
i dont use fill lights etc
i am now getting into using sky materials as lights now for outside the window
i dont use spheres though as it takes up rendering time
hope this makes sense

[FusionDG]

Originally posted by _dd
i always use a light where there is a light, treating it like real world
it seems to work
i dont use fill lights etc
i am now getting into using sky materials as lights now for outside the window
i dont use spheres though as it takes up rendering time
hope this makes sense Yes, that makes sense, although it is easy to go beyond that sometimes with computer software instead of a camera. :)

Though I have also heard others say to treat the scene like photography and therefore to use “soft lights” etc.., to fill when natural light doesn’t give you the effect you’re after. That makes sense to me also. I guess I’m easily swayed :D

So for the light source, are you using a light object or a luminant material applied to an object?

thx,
Paul

[davedavidson]

light maps yes faking hdr or wat ever
i like treating each scene as i would if i was to build it in real life so this helps me set up the lighting better in my head anyway
if its too dark turn up my dimmer switch on the wall or add a table lamp, etc etc

[CGTalk Moderation]

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.