Author Topic: fireflies  (Read 5291 times)

2015-10-31, 05:24:21

3di

  • Active Users
  • **
  • Posts: 128
    • View Profile
I have a scene which seems to be producing a lot of fireflies on black surfaces.  They appear to be mostly in the indirect and reflection passes.

The scene is a pretty large bar/restaurant with high dark (almost black) ceilings and walls, with a lot of reflective/refractive surfaces such as large windows and many wine glasses etc (thin material for the windows/panes, and normal glass material for the glasses).  I've tried increasing ga vs aa, but this needs to be pushed way above the recommended 64 in order to make any substantial difference (over 200), but this makes the rendering time very long per pass.  The direct light pass is fine at 16 samples per gi pass, so I decrease this to maintain that figure when increasing the GI vs AA.

I've noticed also that my ray/s is only at about 700,000, I can increase this by upping the light sample multiplyer, but i'm not sure if that's necessary if my direct light pass is already fine?

I'm using path tracing and uhd cache.

I have 246 light groups, many of which are coronalight materials, but there are also approx 100 or so corona light objects (disc), and have checked for albedo issues (the only bright white sections are the reflective/refractive objects, nothing red)

Also much of the geometry was bought in from sketchup..not sure if this should matter?

Any advice?

2015-10-31, 09:48:44
Reply #1

atelieryork

  • Active Users
  • **
  • Posts: 283
    • View Profile
    • Atelier York
Which version of corona? In 1.2.1 there were issues with disc lights that might be causing issues.

Best thing is to post some region renders or crops etc. so we can see.
Alex York
Atelier York
www.atelieryork.co.uk
max 2016 sp1, corona 1.3 final, win 8.1. pro

2015-10-31, 10:11:20
Reply #2

romullus

  • Global Moderator
  • Active Users
  • ****
  • Posts: 7533
  • Let's move this topic, shall we?
    • View Profile
    • My Models
You're talking some scary numbers here, like 200 for GIvsAA and 16 for LSM. Would you mind to post screens of your settings. Example of rendered image wouldn't hurt too. Lots of fireflies sources may be hidden in materials, especially reflectives and refractives. Lovering MSI may be quite efficient in fighting those little nasty bugs, did you try that?
I'm not Corona Team member. Everything i say, is my personal opinion only.
My Models | My Videos | My Pictures

2015-11-02, 02:46:09
Reply #3

3di

  • Active Users
  • **
  • Posts: 128
    • View Profile
I seem to have lessened the problem a little by reducing glass to reflection/refraction level 0.7.  But I still have to wait approx 600 passes for the fireflies to almost disappear.

Here's a screenshot at 100 passes, you can see the bright pixels scattered around in the dark areas, and also on the light shades on the left, and a bit of the woodwork.

I've used settings of gi vs aa = 16, and light samples multiplier 1.



photoupload

2015-11-02, 04:03:19
Reply #4

Juraj

  • Active Users
  • **
  • Posts: 4314
    • View Profile
    • studio website
The high amount of lights (not groups) indicate you might have used quite complicated geometry for the lights ?

In any case, this scene combines two complicated scenarios (lot of glossy surfaces in GI heavy environment for one, and lot's of lights for two). Both of those are tough to sample, but combined, yeah, nightmare.
I don't see as much fireflies as I see noise, and sometimes yes, 600+ passes (at default settings) it's simply what it takes.

HD success rate is also poor, but this can't be adjusted in any way currently but it's what simply happens in complicated scenes, I battle the same issues.

Also, you probably don't need portals, they're quite gimmicky imho in Corona and often that not, can complicate your situation. In scene like this, 99perc. of light is from inside, you don't need any portals.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2015-11-02, 05:00:44
Reply #5

Christa Noel

  • Active Users
  • **
  • Posts: 910
  • God bless us everyone
    • View Profile
    • dionch.studio
where is the fireflies? i cant find it.. can we see the 600pass picture?
I experienced heavy noise problem on alpha 7 build scece where there was so many lights in a big room and with some kind of .ies. but non complex lights geometry. but ies shouldn't be a problem again right now.
juraj has simply explained it well, hundred lights + many high gloss with bump object in a room is pretty heavy, but more passes to wait is worth to reduce noise for final result.
but could it be caused by portal? portal is noise solution but not for that kind of scene where there is so many light inside the room.

2015-11-02, 07:21:26
Reply #6

3di

  • Active Users
  • **
  • Posts: 128
    • View Profile
Here is another of the same scene, outdoor this time using an hdr for lighting, and also corona rectangular lights off screen.  The fireflies seem to keep adding as it renders.

You have to zoom in to see them.



photo hosting
« Last Edit: 2015-11-02, 07:32:19 by 3di »

2015-11-02, 12:40:19
Reply #7

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 11208
  • Marcin
    • View Profile
I think the noise is caused by:
1. High amount of lights, which probably shouldn't be lights at all:
How is the yellow textured bar counter material made? Is it a LightMtl? Please see: https://coronarenderer.freshdesk.com/support/solutions/articles/5000515603
Generally it is best to optimize the lights as much as possible. So for example instead of assigning a LightMtl to some complex geometry like curvy lightbulbs or lightbulb filaments (yeah, I have already seen scenes like this...) you can just assign them LightMtl with emission off, or self illumination, and add a simple CoronaLight object in the place where light has to be cast.

2. The "light through solid glass" bug. See: https://coronarenderer.freshdesk.com/support/solutions/articles/5000516180
The easiest/fastest solution would be to use thin glass instead of solid.

3. Reflective caustics - light bouncing off reflective surfaces. Solutions would include:
-decreasing MSI (to 10? 5? it shouldn't be a problem as it is a pretty dark scene)
-using rayswitch to decrease reflectivity visible to GI for the glossy materials - so plugging rayswitch map into reflecitivty slot and having a black/darker color for reflections
Marcin Miodek | chaos-corona.com
Chaos Corona Support Team Lead | contact us

2015-11-04, 17:08:04
Reply #8

3di

  • Active Users
  • **
  • Posts: 128
    • View Profile
Thanks for your replies.

The problem with using light objects is that my scene contains over 250 lights, so i've had to use meshes and add light material and then attach them together where possible to come back under the 250 light groups to avoid the engine going into the lower quality render mode.

I did try reducing the msi, but it made no difference unless I took it all the way down to 0.1.

I'll have a look at rayswitching......would this have the same effect as using a darker colour in the reflection colour?

By the way, is there any way to calculate the ideal settings for the aa vs gi and light samples multiplier from the information on the VFB?  I'm currently having the problem where a scene is taking over 20 hours for indirect noise to start cleaning up.  I'd like to calculate the best settings without trial and error if possible.....particuarly as I'm hoping to send off to a render farm, so I wont have the luxury of trial and error.

Cheers

Michael

2015-11-04, 17:37:39
Reply #9

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 11208
  • Marcin
    • View Profile
I did try reducing the msi, but it made no difference unless I took it all the way down to 0.1.
It should not introduce any visual changes, but should slightly speed up rendering - that's the point.

Quote
I'll have a look at rayswitching......would this have the same effect as using a darker colour in the reflection colour?
Yes, exactly, only it won't be visible directly (you will still see the original reflectivity, the darker one will be only visible to GI).

Quote
By the way, is there any way to calculate the ideal settings for the aa vs gi and light samples multiplier from the information on the VFB?  I'm currently having the problem where a scene is taking over 20 hours for indirect noise to start cleaning up.  I'd like to calculate the best settings without trial and error if possible.....particuarly as I'm hoping to send off to a render farm, so I wont have the luxury of trial and error.
Sorry, but I think only this:
https://coronarenderer.freshdesk.com/support/solutions/articles/5000516731
This is too scene-dependent.

Michael
[/quote]
Marcin Miodek | chaos-corona.com
Chaos Corona Support Team Lead | contact us

2015-11-04, 17:59:33
Reply #10

3di

  • Active Users
  • **
  • Posts: 128
    • View Profile
Thanks Maru, my render is currently on 2003 passes, and the noise is sloooowly disappearing in the indirect pass.

Will the noise dissapear in the same number of passes at lower resolution as they will at higher resolution?  Or will different resolutions require different gi vs aa and lsm settings?

Also, is there any correlation between noise and rays per second/hd success?

Cheers

Michael.