Author Topic: Strange doubling of pass numbers  (Read 684 times)

2023-03-05, 20:07:56

archimi

  • Active Users
  • **
  • Posts: 37
    • View Profile
hello,

while wanting to optimize my calculations I noticed a phenomenon which seems strange to me:
Here is a very simple scene (teapots, of course !) with default settings.
I checked "direct visibility override" and I put a solid gray. I put a noise limit at 4%. And if the gray is lower than 157, my image will take 10 passes and if the gray is higher than 157, it will only take 5 passes. That's twice the time!
it is more obvious with a white or black background.
It's normal ?

3ds 2023 et Corona 9 (hotfix 1)

« Last Edit: 2023-03-05, 20:21:41 by archimi »

2023-03-05, 20:54:10
Reply #1

romullus

  • Global Moderator
  • Active Users
  • ****
  • Posts: 8856
  • Let's move this topic, shall we?
    • View Profile
    • My Models
I guess you're using noise level as your render stop condition, right? Thing is that Corona calculates noise slightly differently with light vs dark background. Let's say you set noise limit to 5% and Corona reach 4,9% noise with light background, but 5,1% with dark background. That means that in the first case the render will stop after 5 passes, but in the second case it needs to continue till it reach set target and since by default the noise level is calculated after each 5 passes, Corona will render those additional passes even if that means that your image now has something like 3% noise. Long story short - it's not the bug. If you need more consistent render times, use other render stopping conditions, like pass, or time limit.
I'm not Corona Team member. Everything i say, is my personal opinion only.
My Models | My Videos | My Pictures

2023-03-06, 09:42:41
Reply #2

archimi

  • Active Users
  • **
  • Posts: 37
    • View Profile
Thank you, Romullus, for your very interesting answer. I didn't know this story of noise calculation every 5 passes.
Usually I do still images and I don't worry too much about the calculation time. I just set the desired noise level.
But here, I have to do an animation. I have several thousand images that require 2 hours of computation. And just by changing slightly the background color, I found myself with a 4 hours calculation! it seems crazy to me, for such a small change.
If I understand correctly, you must avoid using noise as a limit for the calculation of the animation?


2023-03-06, 09:54:56
Reply #3

romullus

  • Global Moderator
  • Active Users
  • ****
  • Posts: 8856
  • Let's move this topic, shall we?
    • View Profile
    • My Models
Pretty much yes, when rendering animation, noise level might not be the most reliable thing to predict your total render time. If you're still want to use noise level, but your target is tethering 5 and 10 passes, what you could do is to change the rate at which noise level is evaluated. By default it is set to 5, but you can change it to any value you want, even to 1, so the noise level will be calculated after each pass. Be aware, that this might slightly increase overall render time, so you might want to do some testing before committing to rendering a thousand frames animation. 
I'm not Corona Team member. Everything i say, is my personal opinion only.
My Models | My Videos | My Pictures

2023-03-06, 10:10:19
Reply #4

archimi

  • Active Users
  • **
  • Posts: 37
    • View Profile
By default it is set to 5, but you can change it to any value you want, even to 1,

just out of curiosity where is the setting for this? I don't think I'll be playing with this, but it may come in handy later.

2023-03-06, 10:31:28
Reply #5

romullus

  • Global Moderator
  • Active Users
  • ****
  • Posts: 8856
  • Let's move this topic, shall we?
    • View Profile
    • My Models
Ah sorry, forgot to mention where to find it. It's called adaptivity recalc. and it is located in render setup>performance>dev/experimental stuff. If you don't see experimental rollout in your performance tab, you need to unhide it first - go to system tab open system settings window and tick on enable dev/debug mode.
I'm not Corona Team member. Everything i say, is my personal opinion only.
My Models | My Videos | My Pictures

2023-03-06, 10:40:46
Reply #6

archimi

  • Active Users
  • **
  • Posts: 37
    • View Profile
thank you for your answer !

2023-03-06, 10:43:47
Reply #7

romullus

  • Global Moderator
  • Active Users
  • ****
  • Posts: 8856
  • Let's move this topic, shall we?
    • View Profile
    • My Models
You're welcome!
I'm not Corona Team member. Everything i say, is my personal opinion only.
My Models | My Videos | My Pictures

2023-03-06, 12:45:21
Reply #8

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5468
    • View Profile
As a note, recalculating adaptivity more frequently than 5 passes will reduce the benefits of using it (which is why we set 5 as an optimal balance). As another note, remember that not everything will double, that is just what is happening here. In this case it takes twice as long because only 5 passes were done. If in a scene needing 50 passes, and extra 5 won't make that much difference, and so on. And of course sometimes it won't need any extra passes from making this change, it just so happens in this scene that it does. All that depends on the scene, as to how many extra passes will be needed for the two different colors in the background - though if it's just direct visibility, that really isn't likely to be much.

I probably wouldn't go adjusting your adaptivity recalculations just for this one scene, I'd say makes more sense to set a suitable pass limit in this case :)
Tom Grimes | chaos-corona.com
Product Manager | contact us