PDA

View Full Version : Radiosity questions - non-inter MC



Amerelium
11-11-2016, 12:37 AM
I was Wondering:

I've learned that I can drastically improve render times by decreasing amounts of rays per evaluation and increasing AA samples (from 96 and 1-8 to 24 to 1-24).
I cannot see differences on stills (except better AA) - are there any negative effects on lighting by doing this?

Also, is there any real point in having more than two indirect bounces? I get the impression more will only make it all look more washed out.

Surrealist.
11-11-2016, 02:27 AM
2 IB is usually enough if you want to get a bounce of color in the scene from surfaces. Generally considered enough. More than that and it increases render time without as much of a result for most situations. There are some cases where you would want more but not usually.

For stills you may or may not notices as much difference between AA and RPE. But it will make a difference for animation sequences. Testing on a scene by scene basis is best. Also increasing the threshold from the default of .01 is a great way to get faster renders. Again adjustable based on scene and detail needed etc. And also more noticeable for animation.

jwiede
11-11-2016, 03:26 AM
I was Wondering:

I've learned that I can drastically improve render times by decreasing amounts of rays per evaluation and increasing AA samples (from 96 and 1-8 to 24 to 1-24).
I cannot see differences on stills (except better AA) - are there any negative effects on lighting by doing this?

Also, is there any real point in having more than two indirect bounces? I get the impression more will only make it all look more washed out.

Off the top of my head:

Among other things, it depends on your light-through-transparency needs -- for penetration through/off n transparent surfaces, you'll need n+1 bounces in relevant bounce limit settings. If you've got light that has to go through a bunch of transparent surfaces to light certain areas, you may need more bounces to make that work (and likewise for camera-sourced rays w.r.t. visibility). For certain unusual cases it can matter quite a bit, but more typically 2-3 is a reasonable limit.

ATM, I can't think of other cases akin to the transparency case which "force" need for higher/specific bounce limit settings. I seem to vaguely recall there's another one that's similar, but it was a really contrived example (IIRC).

The "lower RPE/higher AA" tradeoff is kind of specific to LW (at least, IME) but is generally viable in LW (for stills -- never tried it for animation, but can think of how it could create/worsen certain anim artifacts). If you go too low with RPE you can get yourself into trouble with certain types of detailed surfaces, but such cases are typically obvious.

Hope that helps!

S0nny
11-11-2016, 04:56 AM
I was Wondering:
I've learned that I can drastically improve render times by decreasing amounts of rays per evaluation and increasing AA samples (from 96 and 1-8 to 24 to 1-24).
I cannot see differences on stills (except better AA) - are there any negative effects on lighting by doing this?


No negative effects. This settings only contribute to the noise in the shading and/or the AA areas, nothing else.

You can consider to increase the shading samples or light samples over 1 though, because there are situations in which the camera samplings or the RPE are not good enough to clean up the noise. Higher values in light samples also are very fast to calculate than AA and more effective in critial areas, even with adaptive samplings.

This is more evident in non-interpolated MC than the interpolated one: in this regard Lw MC non interpolated have similarities to Arnold samplings, once you understand what's causing the noise it's just matter of increase the right samplings (Shading/Lights/Camera) or some rpe to get a clean results. And wait longer the render to finish.



Also, is there any real point in having more than two indirect bounces? I get the impression more will only make it all look more washed out.
[/QUOTE]

In my experience anything over 4 bounces in Lw is a total waste of time and resources. The more the bounces the more the effect became subtle, you can get pretty much the same result as 10 bounces with two clicks in post or in comp. Really it doesn't worth.

But I'd say it's different from 2 to 4 though, for interiors and some occluded areas in the scene you'll get more natural color bleeding and shadows, for me it still worth the rendertime because it doesn't add up too much.

Btw everything is very scene-dependent of course, but this are a good rules of thumb to start with.

Amerelium
11-11-2016, 05:19 AM
I'm doing 3000+ frame animations, primarily outside environments.

Like this:

https://vimeo.com/184041325

https://youtu.be/8VfGlGeErEI

So the rays per evaluation thing is basically about removing noise then? Since high AA / low RpE render takes about 2/3 the time of my original setup, I'll stick to that then...

S0nny
11-11-2016, 06:19 AM
I'm doing 3000+ frame animations, primarily outside environments.

Like this:

https://vimeo.com/184041325

https://youtu.be/8VfGlGeErEI

So the rays per evaluation thing is basically about removing noise then? Since high AA / low RpE render takes about 2/3 the time of my original setup, I'll stick to that then...

You shouldn't have problem in your animation using low rpe and non interpolated montecarlo, there are some sequences that will render pretty fast actually, but without the headache of the interpolation problems.

RPE in this case (Brute Force MC) affects the noise because there's no interpolation between the sampled areas.

You can think the interpolation method as a blur filter on top of the GI noise: in non interpolated you'll see just the pure noise, in interpolated you see an attempt to clean up the noise by the algorithm (the interpolation). It's not the way it works, but maybe it helps.

The noise affect directly the interpolation, bad noise produce more irregularities in the interpolation. If the noise is really bad (because of the low RPEs), you'll get splotches. If the noise is very fine (because high RPEs) chances are you'll get a much cleaner and uniform interpolated result.
This is also why RPE don't affect camera samplings in interpolated.

Amerelium
11-11-2016, 07:55 AM
Let me rephrase:

If I do a render with 10 rays per evaluation, and then the same one with 1000 rays per evaluation, will there be other differences besides noise level?

And again; non-interpolated

S0nny
11-11-2016, 08:04 AM
If I do a render with 10 rays per evaluation, and then the same one with 1000 rays per evaluation, will there be other differences besides noise level?


As I said, no other difference. The explaination was about why in non interpolated this controls the noise, and why in interpolated it controls the splotches.

Amerelium
11-12-2016, 02:32 AM
Thanks - just want to make sure the lighting itself isn't made worse beyond what can be fixed with adaptive AA.

I guess this only works so far though? If I use only 1 ray, that means light bounce in only one direction, right?

S0nny
11-12-2016, 03:25 AM
Thanks - just want to make sure the lighting itself isn't made worse beyond what can be fixed with adaptive AA.

That's the way it works in Lw.



I guess this only works so far though? If I use only 1 ray, that means light bounce in only one direction, right?

I don't think it's technically correct, because it about the probability that the ray will hit the right spot. The less the rays, the less this probability, so you get more noise.
Bounces setting controls a different thing.
In the 1 ray only scenario it's possible that it's too limited to be cleaned out by AA only, but I don't think theres a right answer to this, it depends in the situation.

For example, there are situations in which I found that AA wasn't enough to clean up the noise in some reflection blur area in some shaders, or was inefficient to clean up the noise from some lights. Using the Shading/Light sample over 1 gave a much better result without add too much AA passes. So for sure the same is possible with very low RPE, like 1.