Tiny-little-testscene-beast - fail to render properly

[Sascha Selent]

Hi everyone.

Okay, here is a tiny little testscene that turns out to be a beast (at least for me). I didn´t make it to render it properly in a good time, struggling for two days now. So maybe someone can give me some hints to bring this beast under control.

Here it is, rendered with my common startup-settings - Irradiance-map with preset "High", LightCache for the secondary bounces, all rendered on 6 nodes (note: I added some contrast in the color-mapping and the GI-settings):



Does not look that difficult, but I fail to bring it under control with an acceptable amount of rendertime. All over a little splotchy, but the problematic areas are the window-frames, especially the tiny corner-areas that are illuminated by secondary bounces only. They are too light with a lack of detail. I think this is the problem - only secondary bounces come close to these areas.

So next I tried to get rid of the splotches by increasing the HSph. subdivs from 50 to 100:



Seems to work, but still some lack of detail around the window-frames.

Switch back to HSph. subdivs to 50 to speedup rendering and tried Detail enhancement in the IRRMAP:



No impact on details.

Next I tried to pimp up the LightCache:



Okay, now some details come up - rendertime too.

Tried to pimp up the LightCache even more - 5000, 0,005 with HSph. subdivs in IRRMAP = 100:



Okay, now it comes to somewhat near perfect. Rendertimes increase significant (remeber, 6 nodes). Now it´s overall a little splotchy, I think I have to increase the HSph. subdivs even more = longer rendertimes.

But I like it more to use the LightCache in World-units instead of Screen-units, because I want the same details further away from the cam. LC 5000, 4cm, Interp. samples = 20:



Quite good. Again allover splotchieness. I try to correct this with higher settings for the IRRMAP: HSph.subdivs = 150, Intep. samples = 30:



I think this is okay. Still some blotches and still a little lack of detail, but when I pimp up everything more, I run into RAM-issues and Max crashes. So that´s my limit.

Okay, last try: upped the HSph. subdivs to 200:



Okay, that´s it! I think the rendertime on a single machine would be around 45 minutes, because the LightCache needs about 25-30 minutes. This is very long for this simple kind of scene. Get the scene here... (Max2008, VRay 1.5 SP2).

Okay, at this point I tried Vlado´s Universal V-Ray settings:



Primay Bounces BruteForce, 8 Subdivs, Secondary Bounces LightCache, 1000 Subdivs, Sample size 0,02, Interp. samples 5.

Rendertime is high but now the window-frames show some problems.

So I tried to pimp up the Universal settings - For the LightCache upped the Subdivs to 3000, lowered the Sample size to 0,01, upped the Interp. samples to 20.:



Better, near to the best I colud get from IRRMAP with the pimped LightCache. But still not perfect. Get the scene here... (Max2008, VRay 1.5 SP2).

Anyway, it seems that everything depends on the LightCache for this scene. I need incredibly high settings to achieve the result I desire.

I also tried to use Skylight Portals to achive a better quality / more detail, but it seems that this leads to longer rendertimes with less quality. Maybe I do something wrong here. In the scene-files there are hidden layers with Skylight Portals - there you can see my tries.

Now I could try to use a different engine for the secondary bounces, e.g. Brute Force, but before I do that I would like to ask you:

How would setup this scene? Would you be able to get a perfect result in short rendertimes? If yes: how do you do it?

Thank you in advance.

Sascha

[peterguthrie]

Hi Sascha,

I had a quick look... I'd stick with vlado's universal settings, but would do that little thing they call lwf. Your white material is very bright, i'd stick it inside a max color correction map (at 0.4545) and i'd probably also lower it to more like 230,230,230. For color mapping, i'd use reinhard, with gamma set to 2.2 and the burn value dependent on the view.

Here is my result with vlado's universal settings but with a noise threshold of 0.005:



On this as well as the white material adjustments I restored the intensity of the sun to 1, used a vrayphysicalcamera at F4, 1/250th & ISO 100, unchecked 'hidden lights' on the vray global switches as I didnt want any vraylightportals. True lwf can make renders look a bit washed out, but i always render full float and just tweak the curves in the vray frame buffer.

Here is my result with IRR map (-5,-1 & Hsph 90, plus DE) and LC at 2000, 0.02:



Hope thats useful,

Peter

[Sascha Selent]

Here is my result with vlado's universal settings but with a noise threshold of 0.005:

Here is my result with IRR map (-5,-1 & Hsph 90, plus DE) and LC at 2000, 0.02:

Thanks man, this looks promising! Can you provide a scene?

Sascha

[peterguthrie]

I would have done, but i'm on max 2009 so it might not be of any use to you? I'll try and answer any questions though!

[Sascha Selent]

I would have done, but i'm on max 2009 so it might not be of any use to you? I'll try and answer any questions though!

I did not install VRay for Max2009 yet, but I can do. So would be helpful unless I cannot reproduce your setup in Max2008.

Thank you.

[peterguthrie]

Your welcome Sascha, here is the max 2009 file. I think i left it at the Irradiance map/LC settings.

[Sascha Selent]

Hm, I think it looses a lot of details...

[Sascha Selent]

Brute Force for the secondary bounces

I think Brute Force does a good job for memory-friendly secondary bounces:

[peterguthrie]

are you rendering for animation or stills?

[Sascha Selent]

are you rendering for animation or stills?

It´s just testing. But it shall be usable for animations too...

...but I´m still on SP2, I will not test for animation prior SP3...

[pengo]

To get more clean GI without artifacts and noisy results:

Personally i'd use not contrast values, it could make more visible what LWF method generally smooth the little noise GI gives.
Use check sample visibility in irrad map options to avoid light leaks in interiors renders
Avoid coplanar faces u have some of those too in your scene (walls onto floor, and ceiling against walls), that can give more noised GI.
If u want better GI around contact areas between objects (like proper shadows indirect light generates) u can increase the dist threshold in irrad map options value to 0.2 or little more.

hope it helps
Sorry my bad english

[peterguthrie]

just noticed the system units of your file were in inches, and the display units in cm.. think that lead to the DE settings I had not really doing anything. Heres a new render, I recreated the scene from scratch by just merging in your geometry so the sun angle might not be exactly the same.

IRR map: -4,0 Hsph 90, dist threshold 0.2 (pengo's tip) DE: 750 world units (I work in mm)
LC: 2000 subdivs
Reinhard color mapping, burn value 0.25 gamma 2.2

otherwise everything default.

[Sascha Selent]

just noticed the system units of your file were in inches, and the display units in cm.. think that lead to the DE settings I had not really doing anything. Heres a new render, I recreated the scene from scratch by just merging in your geometry so the sun angle might not be exactly the same.

IRR map: -4,0 Hsph 90, dist threshold 0.2 (pengo's tip) DE: 750 world units (I work in mm)
LC: 2000 subdivs
Reinhard color mapping, burn value 0.25 gamma 2.2

otherwise everything default.

YES, that´s it. Can you post this scene too?

Thank you!

[Sascha Selent]

Personally i'd use not contrast values, it could make more visible what LWF method generally smooth the little noise GI gives.
Use check sample visibility in irrad map options to avoid light leaks in interiors renders
Avoid coplanar faces u have some of those too in your scene (walls onto floor, and ceiling against walls), that can give more noised GI.
If u want better GI around contact areas between objects (like proper shadows indirect light generates) u can increase the dist threshold in irrad map options value to 0.2 or little more.

Okay, thank you very much. I didn´t know that non-visible complanar faces would have impact on the scene. Visible complanar faces do mess up the GI completely, I know.

[peterguthrie]

here you go. Its still not perfect but not bad for 2mins maybe?