Author Topic: Stubborn Direct Light Noise  (Read 3108 times)

2015-09-18, 10:51:15

Dippndots

  • Active Users
  • **
  • Posts: 296
  • Alex Fagan Co-Founder at The Faction
    • View Profile
    • The Faction
I'm having real trouble cleaning up the direct noise in the below image. It rendered 200 passes, with 32AAvGI, and 16LSM. It took over 2 hours on a dual xeon blade server, and as you can see there is still noise from the outside light (HDRI) and the light in the right top corner, which is a cylindrical corona light acting like a fluorescent tube. It sits inside a "thin" clear box.

As you can see the other light sources in the scene are noise-free. Indirect light noise clears up really quickly in comparison as well.

What am I missing here?


2015-09-18, 12:58:24
Reply #1

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12768
  • Marcin
    • View Profile
There are some informations from the underground that soon there might be a new version of Corona with greatly improved direct sampling, just a rumor... ;)

LSM set to 16 is an overkill. You are casting 32*16=512 direct light samples / pixel / pass! Using a lot more direct than GI samples may be required, but then you should probably decrease GI samples. So you can for example try GIvsAA@4 and leave LSM@16. This way the ratio between GI and direct quality should remain the same, but you will get a lot better per pass performance and much more pixels will be sampled.

Another idea is to use buckets here as there are some large flat solid areas in your render.

Do you have portals in your windows? You could also try placing one in that "hole" through which the noisy top-right corner light is shining. Not sure if this will help, though.

It sits inside a "thin" clear box.
If there is a light inside a reflective/refractive tube, then this could be casting some reflective/refractive caustics. You could disable them by nesting the glass material inside a rayswitch material and clearing the GI slot to make it completely invisible to lights (it will also stop casting any kind of shadows).
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2015-09-18, 13:08:36
Reply #2

Dippndots

  • Active Users
  • **
  • Posts: 296
  • Alex Fagan Co-Founder at The Faction
    • View Profile
    • The Faction
I like the sound of that rumor!

I'll try those settings when I get back from lunch and I'll send over the scene with as much geo as I can.

There are portals in the windows, ill try adding one to the gap in the wall too.

By "thin" I meant it's not refractive, I'll try the rayswitch though.

2015-09-18, 15:21:30
Reply #3

Dippndots

  • Active Users
  • **
  • Posts: 296
  • Alex Fagan Co-Founder at The Faction
    • View Profile
    • The Faction
Putting the GI down certainly makes it faster per pass,but it just moves the numbers around, I need put in 1400 passes to get similar direct light noise as the original settings, with no decrease in render time as a result. Putting a portal doesn't change anything either unfortunately. Using the rayswitch made the render times 5x slower which was weird.

I'm just waiting for the scene to finish uploading, then I'll send it through.

2015-09-22, 14:17:08
Reply #4

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12768
  • Marcin
    • View Profile
Hey, check out the newest daily build, it _should_ be helpful here:
https://coronarenderer.freshdesk.com/support/solutions/articles/5000570015
(just remember to backup your scene before using it)
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2015-09-28, 14:42:02
Reply #5

Dippndots

  • Active Users
  • **
  • Posts: 296
  • Alex Fagan Co-Founder at The Faction
    • View Profile
    • The Faction
Hey Maru, just got back in the office and am playing around with it today. It definitely looks like that daily-build has improved the direct light noise, there's a little more indirect light noise, but I'm re-balancing settings and should have a good render by the end of the day. Already getting similar results in 1:20-1:30 instead of 2hrs.