Spheres scene challenge

Post your tests, experiments and unfinished renderings here.
User avatar
Dade
Developer
Developer
Posts: 5672
Joined: Mon Dec 04, 2017 8:36 pm
Location: Italy

Re: Spheres scene challenge

Post by Dade »

epilectrolytics wrote: Mon Sep 23, 2019 7:11 pm
Dade wrote: Mon Sep 23, 2019 5:45 pm It sounds like a wrong world volume setup :?:
There is a fog box around the room, the camera and emitters are outside and no world volume defined.
When I define one and assign it as exterior to the fog box and the emitters, there is no change in rendering, still no incoming beam and the caustic beams are too bright.

(I think @FarbigeWelt has somewhere tested BiDir against Hybrid with lasers and volumes and found similar inconsistencies.)
If the box is made of NULL material, it may not work. Just use the world volume and remove the NULL material box :idea:
Support LuxCoreRender project with salts and bounties
kintuX
Posts: 809
Joined: Wed Jan 10, 2018 2:37 am

Re: Spheres scene challenge

Post by kintuX »

A side note from observing this scene.
With a simple test, to again show not to expect too much out of PT vs BiDir... no matter how many tricks or time one invests, IME, lack of precision and feature richness is constantly limiting the creation of effortless beauty.

"Laser on. A matte plane in a Fog."

Compute Unit: CPU
Halt Time: 30s

PT
LaserFog_PT.jpg

BiDir
LaserFog_BiDir.jpg

Also, from a quick look into details it seems to me there are quite a few differences between old Classic and New Core without any change to the scene :? IDK, just looks like too much time went into circling around the hot potato... ;)
epilectrolytics
Donor
Donor
Posts: 790
Joined: Thu Oct 04, 2018 6:06 am

Re: Spheres scene challenge

Post by epilectrolytics »

Dade wrote: Mon Sep 23, 2019 8:17 pm If the box is made of NULL material, it may not work. Just use the world volume and remove the NULL material box :idea:
Yep, that worked (for the incoming beam).
With world volume, PathOCL Metro + PGI and light tracing I could trace the rays completely.
Still volume caustics per light tracing are too bright and reflections on glass bodies are also missing
(Simple glass without dispersion for testing).
Screenshot.jpg
User avatar
Noah_
Posts: 26
Joined: Tue Jan 28, 2020 2:44 am

Re: Spheres scene challenge

Post by Noah_ »

epilectrolytics wrote: Tue Sep 24, 2019 7:38 am
Dade wrote: Mon Sep 23, 2019 8:17 pm If the box is made of NULL material, it may not work. Just use the world volume and remove the NULL material box :idea:
Yep, that worked (for the incoming beam).
With world volume, PathOCL Metro + PGI and light tracing I could trace the rays completely.
Still volume caustics per light tracing are too bright and reflections on glass bodies are also missing
(Simple glass without dispersion for testing).
Screenshot.jpg
Could you share that scene please.
Full Hybrid Back/Forward path tracing - I don't like clamping - Bidir it's great but very slow - I have 4096 MiB de Vram and I want to cry.
epilectrolytics
Donor
Donor
Posts: 790
Joined: Thu Oct 04, 2018 6:06 am

Re: Spheres scene challenge

Post by epilectrolytics »

I didn't save that one, but was able to recreate it for v2.3beta on Blender 2.81a.
The "incoming beam" directly from the IES light is not rendered by light tracing (because no caustic before going through the first sphere) and therefore extremely slow to converge.
I did never finish this render but it will need several thousand samples.
Best to use PathOCL with Sobol sampling (better than Metropolis as I just discovered).
The engines have been changed in the mean time but volumetric caustics from light tracing are still too bright IMHO.
Attachments
Luxcore spheres blend 3.zip
(608.02 KiB) Downloaded 236 times
User avatar
Noah_
Posts: 26
Joined: Tue Jan 28, 2020 2:44 am

Re: Spheres scene challenge

Post by Noah_ »

epilectrolytics wrote: Sun Feb 09, 2020 10:18 am I didn't save that one, but was able to recreate it for v2.3beta on Blender 2.81a.
The "incoming beam" directly from the IES light is not rendered by light tracing (because no caustic before going through the first sphere) and therefore extremely slow to converge.
I did never finish this render but it will need several thousand samples.
Best to use PathOCL with Sobol sampling (better than Metropolis as I just discovered).
The engines have been changed in the mean time but volumetric caustics from light tracing are still too bright IMHO.
Wow Thank you very much for taking the time to reconfigure the scene.

Although I have a doubt, why is the mirror material mixed with metal?
Full Hybrid Back/Forward path tracing - I don't like clamping - Bidir it's great but very slow - I have 4096 MiB de Vram and I want to cry.
epilectrolytics
Donor
Donor
Posts: 790
Joined: Thu Oct 04, 2018 6:06 am

Re: Spheres scene challenge

Post by epilectrolytics »

Noah_ wrote: Sun Feb 09, 2020 5:33 pm why is the mirror material mixed with metal?
Mirror material is too perfect for my taste so I mixed in a reflective material with a little roughness to create a bit of a light halo on the mirror surface (did not turn out very well).
User avatar
Noah_
Posts: 26
Joined: Tue Jan 28, 2020 2:44 am

Re: Spheres scene challenge

Post by Noah_ »

Barely 5000 samples :?

Image
Full Hybrid Back/Forward path tracing - I don't like clamping - Bidir it's great but very slow - I have 4096 MiB de Vram and I want to cry.
epilectrolytics
Donor
Donor
Posts: 790
Joined: Thu Oct 04, 2018 6:06 am

Re: Spheres scene challenge

Post by epilectrolytics »

Yep, while light tracing is faster than BiDir, path tracing (of the volumetric scattering from the IES light) is way slower than BiDir so that it makes no sense to render it this way.

I wonder why adaptive sampling (with Sobol) is not catching this.

I think there is already an implementation for user defined sampling (brush? area?) in LuxCoreUI but not yet in Blender which would definitely help with this scene.
epilectrolytics
Donor
Donor
Posts: 790
Joined: Thu Oct 04, 2018 6:06 am

Re: Spheres scene challenge

Post by epilectrolytics »

Noah_ wrote: Tue Feb 11, 2020 1:18 am Barely 5000 samples :?
It just dawned on me that the problem with the incoming light could be easily solved by putting a glass pane in front of the emitting plane, outside of the screen.
Then it will be a caustic beam like the rest and rendered quickly with light tracing :idea:
Post Reply