Metropolis SDS problem?
Forum rules
Please upload a testscene that allows developers to reproduce the problem, and attach some images.
Please upload a testscene that allows developers to reproduce the problem, and attach some images.
Re: Metropolis SDS problem?
The smaller radius did not work out, even after 20 000 passes, the non VM looks way better.
VM 20k passes.
VM 20k passes.
-
- Donor
- Posts: 790
- Joined: Thu Oct 04, 2018 6:06 am
Re: Metropolis SDS problem?
You're right, that doesn't work and I think it's a bug because the light gets ever brighter the longer the rendering which shouldn't be the case.
Try random sampler instead of metropolis, there the lighting is correct (but the SDS paths get problematic again...)
Thanks for joining in and helping with the testing!
Yes, finally luxcoreui works on my Mac, Woohoo!
There are problems with the retina resolution though (everything scaled up x2) and it's way darker than the render view in blender.
Lots of new stuff to figure out!
I have not yet found what the best approach to the SDS problem is, this will take longer (as in longer renderings because none of the fast solutions are working yet).
So it will take more time before I come back with a conclusion, but if the BiDirVM with metropolis sampling did not have the light overcast issue it would be the solution!
Re: Metropolis SDS problem?
There is an option to scale the UI text up (Screen -> UI Font Scale), but there's no such option for the rendered image I think.epilectrolytics wrote: ↑Thu Oct 11, 2018 3:43 pm There are problems with the retina resolution though (everything scaled up x2) and it's way darker than the render view in blender.
Do you have altered the Blender color management settings a lot?
They can't be replicated by luxcoreui, so to remove differences, try to reset them to defaults.
-
- Donor
- Posts: 790
- Joined: Thu Oct 04, 2018 6:06 am
Re: Metropolis SDS problem?
No, I didn't touch those.
Currently I'm trying border (crop) resolution renders with the slower engines/samplers intending to get something like a reference image.
That I'm doing from blender and will get back exploring luxcoreui later.
Also I can deactivate Mac retina scaling (text becomes very small on 2880x1800 @ 15"), that should solve the scaling issue.
Thanks for all the useful advice!
Re: Metropolis SDS problem?
Make sure you have turned off automatic tonemapping. Could explain the brightness differences.
-
- Donor
- Posts: 790
- Joined: Thu Oct 04, 2018 6:06 am
Re: Metropolis SDS problem?
Ah ok, that'll be it.
I was just wondering why my border rendering is darker than the full sized one and would probably figured out that it is the automatic tone mapping soon
So what kind of tone mapping does luxcoreui do internally?
I was just wondering why my border rendering is darker than the full sized one and would probably figured out that it is the automatic tone mapping soon
So what kind of tone mapping does luxcoreui do internally?
Re: Metropolis SDS problem?
Whatever was set in the config file.
There's a window for these settings in luxcoreui, it's an imagepipeline setting.
There's a window for these settings in luxcoreui, it's an imagepipeline setting.
Re: Metropolis SDS problem?
Tuning the start radius even smaller greatly increases the time when you start seeing VM bad side.
What can be done is 20+ png renders that you can blend together in photoshop.
visibilitymap.enable = 0 (was 1)
Shadow rays = 2 (was -1)
replaced blender exported exr with original hdr environment map file.
Subdivided cube around 3M triangles and added some noise around 0.00005 m.
Moved camera.
light.maxdepth = 256 (was 128)
path.maxdepth = 256 (was 128)
removed line "path.russianroulette.depth = 64"
removed line "path.russianroulette.cap = 0.98"
sampler.metropolis.largesteprate = 0.1 (unchanged)
sampler.metropolis.maxconsecutivereject = 2048 (was 32768)
sampler.metropolis.imagemutationrate = 0.1 (was 0.0055)
lightstrategy.type = "UNIFORM" (unchanged)
bidirvm.lightpath.count = 16000 (unchanged)
bidirvm.startradius.scale = 0.0001 (was 0.0003)
bidirvm.alpha = 0.9999 (was 0.998)
film.filter.width = 4 (was 1.2)
What can be done is 20+ png renders that you can blend together in photoshop.
visibilitymap.enable = 0 (was 1)
Shadow rays = 2 (was -1)
replaced blender exported exr with original hdr environment map file.
Subdivided cube around 3M triangles and added some noise around 0.00005 m.
Moved camera.
light.maxdepth = 256 (was 128)
path.maxdepth = 256 (was 128)
removed line "path.russianroulette.depth = 64"
removed line "path.russianroulette.cap = 0.98"
sampler.metropolis.largesteprate = 0.1 (unchanged)
sampler.metropolis.maxconsecutivereject = 2048 (was 32768)
sampler.metropolis.imagemutationrate = 0.1 (was 0.0055)
lightstrategy.type = "UNIFORM" (unchanged)
bidirvm.lightpath.count = 16000 (unchanged)
bidirvm.startradius.scale = 0.0001 (was 0.0003)
bidirvm.alpha = 0.9999 (was 0.998)
film.filter.width = 4 (was 1.2)
-
- Donor
- Posts: 790
- Joined: Thu Oct 04, 2018 6:06 am
Re: Metropolis SDS problem?
Great, thanks for sharing your settings too!
I also noticed that with smaller radii the caustics tend to appear later.
But it would be guesswork when they are "right" which is not the point in using an unbiased renderer.
As long as there is this energy conservation problem I'll stay away from BiDirVM+Metropolis (but I like your high-key shot too).
Instead I used BiDirVM+random sampler with my original textured file @ 4000 samples (2.5h,not clean, but close, maybe 10000 + denoise would do).
This has the right light balance but because clamping is not yet working in BiDirVM there are fireflies in the glass.
The image pipeline tab in luxcoreui shows no settings but only "image pipeline#0" as if to choose when different versions were defined.
Also in the window tab where the settings are I can't find anything regarding tone-mapping.
But now with a fixed value (instead of automatic) exported from blender it corresponds correctly.
Re: Metropolis SDS problem?
Does the denoiser even work with BidirVM, have you tested it?epilectrolytics wrote: ↑Thu Oct 11, 2018 9:11 pm (2.5h,not clean, but close, maybe 10000 + denoise would do).
By the way, if it works, the denoiser has a "Remove Fireflies" option.
In my tests it seemed to fail when too many outliers where clustered together (makes sense), but you can give it a try.