Page 9 of 23

Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)

Posted: Wed Jun 26, 2019 10:20 pm
by provisory
This stuff is starting to be very good!

A small problem:
Volumetric light isn't visible before a specular bounce with hybrid, however after it it's much brighter than with traditional BiDir.

Spot Study and Lenses Array

Posted: Thu Jun 27, 2019 6:10 am
by FarbigeWelt
openCL light trace with indir. cache render scattering in the room less than Bidir. Spots are comparable and laser is visible in mirror.
Spot Study openCL light trace, indir cache 1906 samples 6m21s OIDN 0.90 gain 0.0005
Spot Study openCL light trace, indir cache 1906 samples 6m21s OIDN 0.90 gain 0.0005
Spot Study BiDir 49 samples 6m12s OIDN 0.90 gain 0.0005
Spot Study BiDir 49 samples 6m12s OIDN 0.90 gain 0.0005
Compared to standard openCL open CL with light trace renders causti sharper, better defined and with more color shades.
Gems disperse openCL light trace 4533 samples 8m24s OIDN 0.90
Gems disperse openCL light trace 4533 samples 8m24s OIDN 0.90
Gems disperse openCL std 4561 samples 9m42s OIDN 0.90
Gems disperse openCL std 4561 samples 9m42s OIDN 0.90
Overall this new Hybrid Back/Forward path tracing is very welcome and good visible improvement.

Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)

Posted: Thu Jun 27, 2019 7:41 am
by Dade
epilectrolytics wrote: Wed Jun 26, 2019 6:30 pm While the concrete scene works again, still my glass cubes won't show sky reflections.
Your scene is working fine here:

ok.jpg

Are you sure to use the latest binaries ?
epilectrolytics wrote: Wed Jun 26, 2019 6:30 pm When hybrid lightpass always uses metropolis sampler, would we need to adjust the settings in certain situations?
In my testing it works just fine in all circumstances and I don't feel like I have to change anything,
but just in case, are they taken from BiDir or are there extra hybrid metropolis settings?
The light tracer metropolis sampler uses default Metropolis parameters, there isn't currently a way to change them but it is not needed anyway.

Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)

Posted: Thu Jun 27, 2019 7:42 am
by provisory
provisory wrote: Wed Jun 26, 2019 10:20 pm Volumetric light isn't visible before a specular bounce with hybrid, however after it it's much brighter than with traditional BiDir.
I tested area light.

Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)

Posted: Thu Jun 27, 2019 7:55 am
by Dade
provisory wrote: Wed Jun 26, 2019 10:20 pm A small problem:
Volumetric light isn't visible before a specular bounce with hybrid, however after it it's much brighter than with traditional BiDir.
It is working fine here (and it is mind blowing fast):

vol.jpg

Are you sure to use the latest binaries ? Can you post a test scene ?

Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)

Posted: Thu Jun 27, 2019 9:05 am
by provisory
Dade wrote: Thu Jun 27, 2019 7:55 am Are you sure to use the latest binaries ? Can you post a test scene ?
I've downloaded the binaries again, but it's the same.

BiDir (sobol):
HybridVolumetricLightTest-bidir.jpg
HybridVolumetricLightTest-bidir.jpg (10.08 KiB) Viewed 5839 times


Hybrid:
HybridVolumetricLightTest-hybrid.jpg
HybridVolumetricLightTest-hybrid.jpg (4.13 KiB) Viewed 5839 times

Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)

Posted: Thu Jun 27, 2019 9:26 am
by epilectrolytics
Dade wrote: Thu Jun 27, 2019 7:41 am Your scene is working fine here:

Are you sure to use the latest binaries ?
Umm, downloaded again and reinstalled and now it works :oops:

Sorry for messing up.

But still there are cases where the sky isn't properly rendered:
(right: BiDir, left: original file with hybrid and PGI)
Screenshot.jpg
The attachment cube3_LuxCore.zip is no longer available
EDIT: Probably not a hybrid path issue: simple PathOCL also does not show the sky, but RTPath does :?

The speed of the combination of PGI and Hybrid, even with volumetric caustics or dispersion, is absolutely mind blowing!

Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)

Posted: Thu Jun 27, 2019 11:27 am
by lacilaci
These results are crazy. :shock:

Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)

Posted: Thu Jun 27, 2019 11:34 am
by Dade
provisory wrote: Thu Jun 27, 2019 9:05 am
Dade wrote: Thu Jun 27, 2019 7:55 am Are you sure to use the latest binaries ? Can you post a test scene ?
I've downloaded the binaries again, but it's the same.
Try to render the scene with normal path tracing: it is all black because it has never been able to render lasers and lasers-like light sources (i.e. area lights with an emission angle of 1 degree like in your test scene) with volumes. It is a known limitation, there from day one of laser/laser-like light source support.

Hybrid can now render the light scattered after the mirror bounce (i.e. the caustic path). Light tracing could be used to fill this particular "hole" in normal path tracing too but it is not one of my top priority at the moment.

Re: Hybrid Back/Forward path tracing (aka BiDir without MIS)

Posted: Thu Jun 27, 2019 11:57 am
by Dade
epilectrolytics wrote: Thu Jun 27, 2019 9:26 am But still there are cases where the sky isn't properly rendered:
(right: BiDir, left: original file with hybrid and PGI)
Screenshot.jpg
cube3_LuxCore.zip

EDIT: Probably not a hybrid path issue: simple PathOCL also does not show the sky, but RTPath does :?
It is the result of a malformed scene: you have not set the world volume. The paths starting from light sources have no volume assigned so the are not scattered in light tracing but they are in eye path tracing (because the camera has the correct volume). This explain the different behavior according the rendering engine (i.e. the direction toward path are traced).

If you set the world volume, you get the same and correct rendering out of all render engines.

P.S. it is like 1000 times faster than BiDir+Metropolis :shock: