Author Topic: Adaptivity intenity slider?  (Read 5428 times)

2016-05-17, 14:33:25

cecofuli

  • Active Users
  • **
  • Posts: 1577
    • View Profile
    • www.francescolegrenzi.com
Hi Ondra,

I'm working on a file, with very, very strong DOF and bokeh effect.
After 10 minute, everything is perfect, but on the reflective metal part, the bokeh is SO noisy =(
Maybe, it's interesting to have a slider, a value to make the adaptivity stronger than now.
The default value is good for the 90% of project, I think. But there are situations, where we want more adaptivity.

what do you think?

2016-05-17, 18:45:49
Reply #1

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12754
  • Marcin
    • View Profile
I don't think there is something as "stronger adaptivity". It is either on or off. You can however change adaptivity recalculation time in devel/debug settings. It is the number of passes after which adaptivity is recalculated. Setting it lower should make adaptivity more precise, but the rendering slightly slower.

You could also experiment with setting this value very high and see how it performs.

Any chance of sharing the scene? :)
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2016-05-17, 20:43:39
Reply #2

cecofuli

  • Active Users
  • **
  • Posts: 1577
    • View Profile
    • www.francescolegrenzi.com
In V-Ray, you can chose how much V-Ray can be "adaptive".
I'm talking about a scene with very strong DOF, big and bright bokeh,
I'm using GI vs AA = 2.
Adaptivity recalculation doesn't change nothing.

I cannot share the scene, but this is a test render at 640x640.

At this resolution, with this simple scene and with an i7 970, it's strange that I have so much noise in the bokeh...
Especially, because the rest of the image is good.
If I want a clean bokeh, we need to wait 6-7 hours.
My idea is to have more uniform noise, in case of strong DOF or Motion blur.
Maybe automatically, Corona can do this: strong noisy effects (glossy, dof, Mblur) more adaptivity.



2016-05-18, 16:31:21
Reply #3

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12754
  • Marcin
    • View Profile
I have already reported something similar on mantis.
Can you try setting adaptivity recalc interval to some higher value? I can see that you are getting 3000+ passes, so I would suggest setting it to 50-100.
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2016-05-18, 16:36:05
Reply #4

cecofuli

  • Active Users
  • **
  • Posts: 1577
    • View Profile
    • www.francescolegrenzi.com
 Adaptivity recalc doesn't change a lot the result.
Yes, 3000+ beaucse strong DOF, soe GI vs AA = 4 or 2 ( I don't  remember exaclty)
With GI vs AA = 16 the situation is worse.

but, I know, this is an extreme situation.
for example, this user had a better result without adaptivity...

https://forum.corona-renderer.com/index.php/topic,11994.0/topicseen.html

2016-05-25, 23:46:22
Reply #5

kumodot

  • Active Users
  • **
  • Posts: 315
  • Corona ADDICTED and Evangelist
    • View Profile
    • Kumodot
I had a "lot of problems" with this exacly behavior on a scene with strong DOF/bokeh... The image is clean in like 30-40 min but the Dof is noisy even after 3-4-5 hours... ;(

     This was the Job...

     I was using Corona 1.2 or 3, i canĀ“t remember now... The "solution" was to render the ZDepth with the DOF ON, and then Crank up the Bockeh/Dof on Post... Add more Dof in Post, cleaned up the beauty noise...
   
     
Your Portuguese is worse than my english.
--------------------------------------------
3DGeneralist/Vfx Supervisor

2016-05-26, 12:29:00
Reply #6

cecofuli

  • Active Users
  • **
  • Posts: 1577
    • View Profile
    • www.francescolegrenzi.com
In Corona 1.4 the situation is better that v1-2 or v1.3, because the new adaptivity.
But, the new adaptive sampling need more improvement, in my opinion.

2016-05-26, 21:23:28
Reply #7

zuliban

  • Active Users
  • **
  • Posts: 56
    • View Profile
im also having the same trouble im not sure if this  a common raytracer problem a small source of light intense  bokeh in glass or metal surfaces 1.3 and 1.4 have the same spot of problems i was thinking 1.4 could resolve this but it did not if i render a crop of the small place it resolves in like 9k passes .


2016-05-26, 21:34:03
Reply #8

cecofuli

  • Active Users
  • **
  • Posts: 1577
    • View Profile
    • www.francescolegrenzi.com
zuliban, did you already try GI  vs AA = 2? Maybe 4 is better, because you have a lot of GI.
But, I'm sure, Corona needs some improvement in this area (adaptive sampling, DOF and bokeh)

2016-05-26, 21:55:37
Reply #9

zuliban

  • Active Users
  • **
  • Posts: 56
    • View Profile
i have tried 16, 64, 2 and 4 , i tried this scene in other renders and they fail in the same areas ,corona cleans them faster, so im sure this is a raytracer limitation or issue but at least it cleans with alot of time, i  just don't want to use post dof it looks fake.  but 1.3 cleans them better than 1.4 with bucket mode why not add bucket mode again with the new adaptivity?
« Last Edit: 2016-05-26, 21:59:19 by zuliban »

2016-05-26, 22:05:50
Reply #10

cecofuli

  • Active Users
  • **
  • Posts: 1577
    • View Profile
    • www.francescolegrenzi.com
I agree with you that Corona works better in this situation that other engine. And I think that yes, it's a very difficult problem to solve.
About bucket, I think that was only an experimental stuff
I think he doesn't  support the new adaptivity and some other new features. But I'm not 100% sure.
We must to ask to Ondra why.

Look at my old threads. I think you will find interesting =)

https://forum.corona-renderer.com/index.php/topic,1380.msg10247.html#msg10247

https://forum.corona-renderer.com/index.php/topic,1448.msg10795.html#msg10795




2016-06-02, 23:42:42
Reply #11

cecofuli

  • Active Users
  • **
  • Posts: 1577
    • View Profile
    • www.francescolegrenzi.com
 I7 970, 17 minutes, at 640x640, with PTS=2,. too much noise =(
PS: there isn't any denoise.


2016-06-03, 04:19:24
Reply #12

Juraj

  • Active Users
  • **
  • Posts: 4762
    • View Profile
    • studio website
Bokeh on shiny specular against strong light was always trouble, even in Vray.

I kind of gave up on such situation and just go post-dof : /
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!

2016-06-03, 14:45:56
Reply #13

cecofuli

  • Active Users
  • **
  • Posts: 1577
    • View Profile
    • www.francescolegrenzi.com
Last time I worked with V-Ray was 3 years ago.
So, I don't know if there are some tricks for optimize this scene.
But, I wanted to try the same Progressive method, as in Corona.
Corona did a better job.
PS: in VRay, the metel for the lamp is too glossy. That's why I have less bokeh effect.
But, the goal was to see the noise in the santa.



2016-06-03, 14:47:48
Reply #14

cecofuli

  • Active Users
  • **
  • Posts: 1577
    • View Profile
    • www.francescolegrenzi.com
But, I would like to have some "color threshold", like in the old Corona 1.3 bucket mode.
I prefer more uniform noise (with more noise of course), instead of perfect wall, but so bad noisy bokeh. =)

2016-06-03, 16:17:19
Reply #15

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12754
  • Marcin
    • View Profile
Hmm... I tried 5-minute renders with different settings, and I cannot find a good solution. :/

The only thing that makes the noise acceptable is highlight clamping set to a very low value like 10 (do not mistake with highlight compression), which obviously also affects the appearance of lights...
(see hc10.jpg)
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2016-06-03, 16:58:00
Reply #16

cecofuli

  • Active Users
  • **
  • Posts: 1577
    • View Profile
    • www.francescolegrenzi.com

2016-06-03, 20:18:05
Reply #17

Ludvik Koutny

  • VIP
  • Active Users
  • ***
  • Posts: 2557
  • Just another user
    • View Profile
    • My Portfolio
Something like "color threshold" is pretty much noise limit. It's almost the same thing. If you want maximal bokeh quality, then try to increase adaptivity recalculation parameter in devel debug. Try something like 25, 50 or even 100, considering you have reached 3500 passes. It will give adaptivity better base to start iterating on.

Anyway, I do not think extreme contrasty bokeh is a solved problem in any renderer, it's simply difficult scenario for regular path tracer to resolve. Why not use denoiser here?

EDIT: nevermind, did not notice Maru has already tried higher recalc interval.

2016-06-03, 21:14:08
Reply #18

antanas

  • Active Users
  • **
  • Posts: 269
  • Hmm ...
    • View Profile
As much as I would want adaptivity to be able to cope with such cases in an efficient way and renderers\hardware in general to be able to render that in a blink of an eye too, I think for now, such effects as the bokeh and dof are better (well if not better then certainly faster) done by using post dof generated by z-depth pass for example by using either vfb+'s dof
, which is quite good for the price and what it is capable of, or using by using lenscare plugin
(Alex's Roman's use of it) which is sometimes even better than the rendered dof and bokeh because of one subtle effect it does and other methods\plugins do not do - not sure how it is called correctly but as I remember it is called aerial lens or aerial aberrations or something like that, which occurs on\between the edges of an out of focus objects and produces some nice distortions on objects visible past them which is usually not possible to do by conventional rendering or at least not with the current generation of renderers (correct me if I'm wrong) - this effect might sound like some minor thing until you see the difference on your own renders ) Of course as always glass\almost transparent objects sitting in front of a camera are a problem with post processing method (which is avoidable at least in some cases and to some degree by setting those material to be not visible in masks thus excluding them from generated z-depth pass completely) but the time savings are tremendous even for stills (which I usually do) and I cannot even imagine how much time one could save doing some animations. That's not completely discussion related but I thought it could be useful for someone interested in doing that stuff in an efficient manner ))

edit: Forgot to mention, lenscare's photoshop plugin's implementation is, sadly, almost completely unusable or at least very hard to use and understand, piece of crap - I use it's after effects version in an old version of Fusion which back then could use after effect's plugins - hope someday, they will make some user friendly and logical photoshop version but right now it is such as I described (
« Last Edit: 2016-06-03, 21:23:54 by antanas »

2016-06-03, 22:05:44
Reply #19

Juraj

  • Active Users
  • **
  • Posts: 4762
    • View Profile
    • studio website
Yeah, Lenscare is beautiful, used it on every (and I did very few sadly...) animation I've done, always satisfied.

VFB+ Dof is also pretty good but sometimes produces some unexpected result (sharp line in foreground shallow dof) although I again suspect this might be because of my incorrect z-depth setup.
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!

2016-06-03, 22:20:13
Reply #20

antanas

  • Active Users
  • **
  • Posts: 269
  • Hmm ...
    • View Profile
Yeah, Lenscare is beautiful, used it on every (and I did very few sadly...) animation I've done, always satisfied.

VFB+ Dof is also pretty good but sometimes produces some unexpected result (sharp line in foreground shallow dof) although I again suspect this might be because of my incorrect z-depth setup.

VFB+'s Dof is improved since 2.6 version - it seems that in the end that goddamned sharp line was indeed a bug\some corona aa incompatibility and not our incorrect z-depth setups so it was fixed in 2.7 or at least I don't see much of it anymore )

2016-06-03, 22:48:31
Reply #21

Juraj

  • Active Users
  • **
  • Posts: 4762
    • View Profile
    • studio website
Yeah, Lenscare is beautiful, used it on every (and I did very few sadly...) animation I've done, always satisfied.

VFB+ Dof is also pretty good but sometimes produces some unexpected result (sharp line in foreground shallow dof) although I again suspect this might be because of my incorrect z-depth setup.

VFB+'s Dof is improved since 2.6 version - it seems that in the end that goddamned sharp line was indeed a bug\some corona aa incompatibility and not our incorrect z-depth setups so it was fixed in 2.7 or at least I don't see much of it anymore )

Ah, didn't mean that :- ) Yes, that is solved, 2.7 is really good. I meant sudden lack of smooth transition, like if you would blur object within its boundary only.
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!

2016-06-04, 14:40:51
Reply #22

antanas

  • Active Users
  • **
  • Posts: 269
  • Hmm ...
    • View Profile
Well, somehow I hadn't run into that particular issue so didn't understand what you mean - could you post some examples of how this particular bug looks - maybe I did run into it but just hadn't noticed that something is wrong or I do something differently in z-depth workflow though I don't see what could it be ))

2016-06-05, 15:34:52
Reply #23

Rotem

  • Active Users
  • **
  • Posts: 64
    • View Profile
Yeah, Lenscare is beautiful, used it on every (and I did very few sadly...) animation I've done, always satisfied.

VFB+ Dof is also pretty good but sometimes produces some unexpected result (sharp line in foreground shallow dof) although I again suspect this might be because of my incorrect z-depth setup.

VFB+'s Dof is improved since 2.6 version - it seems that in the end that goddamned sharp line was indeed a bug\some corona aa incompatibility and not our incorrect z-depth setups so it was fixed in 2.7 or at least I don't see much of it anymore )

Ah, didn't mean that :- ) Yes, that is solved, 2.7 is really good. I meant sudden lack of smooth transition, like if you would blur object within its boundary only.

That sounds like a typical artifact of setting the wrong "direction" for the depth buffer (white is near vs white is far). Could that be the case?

2016-06-05, 15:52:07
Reply #24

Juraj

  • Active Users
  • **
  • Posts: 4762
    • View Profile
    • studio website
Hopefully might be, will revisit this scene to do quick test :- ).
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!