Hi there

I've been struggling with a ‘simple’ scene of a gallery space that is lit by a portal light from a skylight and rendered with PBR. I've simplified the scene so the art works and panels dividing the building are omitted.

I've tried various different mantra settings suggested in other discussions and also some tutorial settings(maybe outdated as the min max indirect rays option is no longer in the standard parameters) but I still can't shift the noise completely and can only get half way there with ridiculously long render times - should an hour a frame or more be normal? FYI The largest amount of noise is on the ceiling.

It could be that I'm asking too much of mantra to light a scene such as this with just a portal light without additional help from other lights. Also I noticed an array of speed ups and denoising after cleaning up intersecting geometry and portal light intersections with geometry (portal lights are circular and rectangular).

I was wondering if anyone could suggest possible directions of exploration for the mantra settings and also does anyone have any tips on cleaning up intersecting geometry.

Thanks

John

Here are basic speed up tips I use with Mantra:

First switch to Physically Based Rendering.
Then increase the tile size from 16 to 36.
While you are on that tab also change the drop down to use all processors, not all processors but one.
Don't drop your Adaptive Noise value so low, you'll just spend all your time at the high sample value. And keep your samples a little close to one another. You have 1 and 20. Try something like 3 and 9 or 4 and 16. For adaptive noise try high values like 0.9 just to see how much faster you can render.

Overall these kinds of scenes challenge any render system. What you really need is an additional pass like a photon pass. Then that information can help define light in nooks and crannys so your main noise value does not have to be so high.

Sorry, I don't have any help on photon maps, I have not used them in Mantra. I have used them in Renderman and Redshift and they help get that clean GI look.

Try the GI light, it will improve the render quality greatly…plenty of docs on how to use it

see attached, 20sec render.
If its a simple room with nothing animated you can get away with manual lighting.

Thanks guys for taking the time to look at this its much appreciated.

I'm ‘chasing the noise’ as described in another post but still I find the only thing that is getting rid of the noise is upping the max ray samples if I'm avoiding upping the pixel samples and still not getting rid of the noise in PBR.

Enivob thanks for the mantra setting tips - what is the behind making the sample levels closer? is it to make sure there are enough samples in the more demanding areas of the scene? On other threads I've noted that the min goes no higher than 1. Am I wrong in thinking upping the minimum will increase rays and noise on open areas receiving enough rays already?

Sidenimjay and Enivob - about adding the photon map GI lights - they should help in the areas not getting enough direct light and are creating too much noise ie. the ceiling and corners. I'm still finding a certain amount of noise that won't go away.

CPB that scene you've created is really very clever and there's a lot I can learn from looking through it thank you.Did you clean up the geometry to make certain there were no mistakes with it or did you notice some problems with it?
The Micro polygonal render is very sharp and quick, it doesn't have the PBR look of bouncing diffused colour but it looks pretty cool , I was intending to add some animated elements in the scene would you use this manual lighting approach for an animated scene or do you think it would get too complex with too many changing shadows and reflections ?

Thanks people

The threshold acts like a switch, not a meter AFAIK. So below the threshold you get MIN samples and above or equal you get MAX samples. When you specify 1 for min and 20 for max you are actually saying use 1 or 20 samples, not 1 to 20 samples. By moving the numbers closer together you can even the spread so to speak. Then you are saying use 4 or 16 samples. That is just my preference, one sample is never enough. You need at least 2-3 minimum.

I have noticed no speed difference in using Micro Polygon over PBR. They seem to render the same speed in my scenes.

CPB's scene is fast but it does not support any shadows. So he is increasing speed by accepting that caveat. Once you drop in a shadow casting object the lights will need to be revised and you will once again be faced with normal Mantra speeds, which are slower than CPB's nice example for solving for a specific shots/scenario.

HUGHSPEERS
Did you clean up the geometry to make certain there were no mistakes with it or did you notice some problems with it?

habit; less nodes, less polys.

If there's animated stuff or complex geometry it becomes a pain to fake bounced light. Much easier to go automatic, but getting rid of that last 20% of noise will cost you a massively disproportionate amount of processing time. To be honest its probably worth looking into renderers that are optimized for this sort of thing.

Enivob
The threshold acts like a switch, not a meter AFAIK. So below the threshold you get MIN samples and above or equal you get MAX samples. When you specify 1 for min and 20 for max you are actually saying use 1 or 20 samples, not 1 to 20 samples. By moving the numbers closer together you can even the spread so to speak. Then you are saying use 4 or 16 samples. That is just my preference, one sample is never enough. You need at least 2-3 minimum.

Enivob thanks for making that clearer. I guess the same applies to the indirect sample limits then, maybe I should play with pushing these values closer together as well.

cpb
If there's animated stuff or complex geometry it becomes a pain to fake bounced light. Much easier to go automatic, but getting rid of that last 20% of noise will cost you a massively disproportionate amount of processing time. To be honest its probably worth looking into renderers that are optimized for this sort of thing.

I have been thinking I need to look for other options, I've been hammering away with this scene in mantra regardless as it's a good learning exercise, I'm hoping the knowledge will sit well when I have an easier scene to render That leads to another few questions - what kind of scenes is mantra PBR optimised for and what is good for these kind of scenes.

Thanks again for your help

Enivob
The threshold acts like a switch, not a meter AFAIK. So below the threshold you get MIN samples and above or equal you get MAX samples. When you specify 1 for min and 20 for max you are actually saying use 1 or 20 samples, not 1 to 20 samples. By moving the numbers closer together you can even the spread so to speak. Then you are saying use 4 or 16 samples. That is just my preference, one sample is never enough. You need at least 2-3 minimum.

Uhm, no? Minimum is the starting point everywhere, so even if an area only “need” 1 ray it will use the minimum. Then based on the threshold it will increase sampling to get rid of noise until either a) the maximum rays is reached or b) you've hit the noise threshold, whichever comes first. If the threshold is set too low it will probably always use the maximum amount of rays all the time, so it takes some fine-tuning depending on the case.

@Skybar, thanks for clearing that up.

What is nice about Redshift is that you can visualize your samples in the render. In these images the bright areas represent areas of noise that ended up using MAX sample count.

For best speed and quality you want to strive for a greyscale representation. Then you are not sending out too many MAX ray counts.

Image Error Threshold Of Images:
0.5 (too high) mostly low sample count used.
0.1 (just right) a mix of low and high sample counts.
0.01 (too low) mostly high sample counts used.

You can do that in mantra too, there are checkboxes for image planes of direct/indirect ray samples.

Thanks again everyone for your input.
I've been researching a few other possible renderers and still finding this interior scene difficult to rid of noise so not just a mantra thing

While looking at renderman I came across a good article that seems very clear about adaptive sampling and using min and max samples. check it out. I guess the values suggested shouldn't correspond but the theory should don't you think ?

https://community.renderman.pixar.com/article/1673/debugging-and-optimization-1.html [community.renderman.pixar.com]

Thanks

John

Yes, mantra is definitely not alone in having hard time with noise when scenes are lit with mostly indirect light sources…

That said there are a couple additional tips I can offer:

1. increase diffuse quality instead adjusting ray variance samples and noise level - if the majority of the noise is coming from diffuse bounce (like the example scene on the original post), you'd want to increase the number of indirect diffuse rays independently of the number of reflection/refraction/sss/direct shadow rays which are not causing noise. You'll find this is particularly helpful for production shots with lots of mixed material types.

2. consider not using portal lights - This really depends on what types of environment map you're using. Currently mantra's portal lights do not importance sample against the assigned environment map, which means that when there are high contrast light sources in the environment map, the render can be noisier compared to regular environment light.

Hi there,
Have a look at the attached render, if it is closer to what you are looking for.

i have cleaned up the portal lights, added a new mantra node and gi light. The noise is not completely gone but it it way better i think. you can up the pixel samples to get less noise but for this render i just used 7.

hope this helps.

It was rendered on my MacBook Pro

Renderman might be better at this:


https://rmanwiki.pixar.com/display/REN/PxrPortalLight [rmanwiki.pixar.com]

https://www.disneyresearch.com/publication/portal-masked-environment/ [disneyresearch.com]

https://s3-us-west-1.amazonaws.com/disneyresearch/wp-content/uploads/20150719230142/Portal-Masked-Environment-Map-Sampling-Paper.pdf [s3-us-west-1.amazonaws.com]