Page 22 of 23
Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)
Posted: Thu Aug 01, 2019 3:00 pm
by epilectrolytics
For testing hardly reproducible stuff like this it would be great to have Hybrid established as accessible engine in LuxCoreUI (HybridCPU & HybridOCL) with partition and glossiness threshold as parameters within the engine editor.
Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)
Posted: Thu Aug 01, 2019 3:58 pm
by Asticles
I have reproduced it, 1000 samples each.
Just reloaded the scene file.
Hybrid + rough glass issue again
Posted: Sun Nov 10, 2019 3:23 pm
by epilectrolytics
Same problem, new test file:
Animation with Hybrid engine and rough glass body.
In every frame the caustics show a different brightness, totally inconsistent.
No caches activated.
Currently I have to use Cycles for animations with rough glass which is a shame really.
Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)
Posted: Mon Nov 18, 2019 10:14 am
by epilectrolytics
Tried latest build of BlendLuxCore because of the solution to a similar issue referenced
here but the problem with caustic brightness inconsistency in case of light tracing & rough glass still persists
The animation from the file in the post above shows the same caustic flickering as before (on my hardware).
Re: Issue with Rough Glass confirmed
Posted: Mon Nov 18, 2019 8:07 pm
by FarbigeWelt
BlendLuxCore latest build_f17e31a
LuxCoreRender latest_ 3e1ff1b
I can confirm the behavior of light tracing in the simple scene of
epilectrolytics is unpredictable
. The brighter, noisy renders happen randomly, see 2. and 3. animated GIF.
Animated GIF, click to view
Standard LuxCoreRender Glass
- simple-glass
Rough LuxCoreRender Glass, first renders
- simple-rough-glass_1
Rough LuxCoreRender Glass, second renders
- simple-rough-glass_2
Code: Select all
frame intensity
5 x 24 x 48 x
15 o 26 x 51 o
17 o 32 x 59 x
19 o 44 x 65 x
23 o 46 xx
frames of 75 type every n frame
14 all 18.7% 5
7 x 9.3% 11
1 xx 1.3% 75
o bright x brighter xx very bright
- o bright: simple_rough_0023
- x brighter: simple_rough_0044
- xx very bright: simple_rough_0046
Re: Issue with Rough Glass confirmed
Posted: Mon Nov 18, 2019 9:11 pm
by Martini
This is what I get when I render the whole sequence overnight using yesterday's fixed automatic build
- Animated PNG, click to view
There is some flickering with inconsistent brightness from frame to frame, but the "explosion" that you guys are seeing is not present in my renders anymore.
- I think this is my brightest frame
FarbigeWelt wrote: ↑Mon Nov 18, 2019 8:07 pm
I can confirm the behavior of light tracing in the simple scene of
epilectrolytics is unpredictable
.
I think this is the best way to put it
FarbigeWelt wrote: ↑Mon Nov 18, 2019 8:07 pm
The brighter, noisy renders happen randomly
Yes, unfortunately I don't think it is always good for animations yet, though I have seen some nice caustic animations on YouTube that don't seem to suffer these issues.
Re: Issue with Rough Glass confirmed
Posted: Mon Nov 18, 2019 10:02 pm
by FarbigeWelt
Martini wrote: ↑Mon Nov 18, 2019 9:11 pm
Yes, unfortunately I don't think it is always good for animations yet, though I have seen some nice caustic animations on YouTube that don't seem to suffer these issues.
Well, the most strange thing is that this sporadic light splashes only happen with Rough Glass. Standard Glass shows no brightening and caustics should be animatable smoothly with current builds.
Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)
Posted: Mon Nov 18, 2019 10:12 pm
by epilectrolytics
Thanks Guys for testing my scene and confirming the issue.
I prefer to use rough glass with minimal roughness over clear glass because it shows better reflections, even though it renders slowly.
Sadly because of the inconsistent brightness issue it is not usable with fast light tracing, even in still images the brightness is mostly wrong when compared with BiDir.
Cycles renders rough glass caustics with no problems
Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)
Posted: Mon Nov 18, 2019 10:32 pm
by Martini
epilectrolytics wrote: ↑Mon Nov 18, 2019 10:12 pm
Thanks Guys for testing my scene and confirming the issue.
No problem!
Renders? Yes. No problems? No problems. Caustics? Well... I didn't see any
Could it be that metropolis used for light tracing needs simply more samples to catch all the paths and produce a predictable result? Also if you are using OpenCL for your eye tracing then you will most likely get a huge sample count discrepancy between eye and light path samples. I had to use CPUPATH and render the thing01 scene I posted to 27648 samples to make sure it had 12288 samples for light paths. And even then it only started to
reduce noise; it didn't completely disappear.
Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)
Posted: Tue Nov 19, 2019 7:59 am
by epilectrolytics
Martini wrote: ↑Mon Nov 18, 2019 10:32 pm
Caustics? Well... I didn't see any
Because of blurry refraction at some distance there are no caustic patterns but rather a uniform illumination on the ground from those caustics.
That's actually a case Cycles can deal with
and you need a comparison w/o caustics to see it:
(GIF from Cycles render, 20 000 samples, undenoised)
Martini wrote: ↑Mon Nov 18, 2019 10:32 pm
Could it be that metropolis used for light tracing needs simply more samples to catch all the paths and produce a predictable result?
The comparison with clear glass @FarbigeWelt made above shows that Metropolis works consistently.
Especially small roughness values should show no big difference once switching to rough glass but LuxCore just cannot handle it currently, there has to be some bug in rough glass or light tracing algorithm to explain our results.