metropolis image mutation rate

Use this forum for general user support and related questions.
Forum rules
Please upload a testscene that allows developers to reproduce the problem, and attach some images.
Post Reply
nocivo
Posts: 17
Joined: Sun Jan 28, 2018 5:33 am
Location: Japan
Contact:

metropolis image mutation rate

Post by nocivo »

Hello
what is it exactly the "image mutation rate" parameter of metropolis sampler? I read on wiki that is "Maximum distance over the image plane for a small mutation ". I cannot find more information about it. I can't understand what influence have on the render. I did some fast test with bidir, changing it to 0%, 1%, 10%, 50% and 100% and it seams to me that changes the grain. Seams to me that it's better using 100% because there is more uniform noise grain. Maybe I have to test more in depth with different scenes and light condition, but first I would like to understand better what it is.

someone can explain better this parameter and how influence the render output?

Thanks
User avatar
B.Y.O.B.
Developer
Developer
Posts: 4146
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: metropolis image mutation rate

Post by B.Y.O.B. »

Usually the rule with render settings is: if you don't know what it is or does, don't touch it and use the default value.
If you want to know the whole story you should read an article about the Metropolis sampler.
To be honest I myself also don't know when it would be good to change this parameter.
User avatar
Dade
Developer
Developer
Posts: 5672
Joined: Mon Dec 04, 2017 8:36 pm
Location: Italy

Re: metropolis image mutation rate

Post by Dade »

nocivo wrote: Tue Apr 10, 2018 9:24 am someone can explain better this parameter and how influence the render output?
Metropolis sampler generate a new sample based on the previous one (trough a "mutation"). There are 2 types of mutations:

1) larges: they are totally random mutations (no relation with the previous sample);

2) smalls: they are based on small changes of the previous sample;

"sampler.metropolis.largesteprate" parameter controls ratio between large and small mutations.

"sampler.metropolis.maxconsecutivereject" parameter controls the max. number of consecutive small mutations.

"sampler.metropolis.imagemutationrate" parameter controls how far a small mutation can move on the image plane.

Try to do the rendering with PATHCPU+Metropolis of a sharp caustics: you will notice that with a small "imagemutationrate" Metropolis will be nearly "trapped" in rendering caustics. If you use a large mutation rate, the samples will be spread more uniformly over the image plane.

So, combined with the "largesteprate", controls how much Metropolis should be "adaptive" or "uniform".

Anyway, it is a lot of technical stuff, quite hard to control. I doubt that it can be of any practical use for most users.
Support LuxCoreRender project with salts and bounties
nocivo
Posts: 17
Joined: Sun Jan 28, 2018 5:33 am
Location: Japan
Contact:

Re: metropolis image mutation rate

Post by nocivo »

Thanks Dade, very clear explanation. Now I understand a little more, thanks to your explanation and my tests. Now I have understand that this setting influence the noise of transparent material like glass (maybe also for sss materials). If you have a scene with many transparent material you can lowing image mutation rate settings and large mutation probability, to have transparent material with less noise. In common situation it's better to leave this parameter as they are, but lowing this setting when you have a lot of transparent material. With bidir, it seams that this settings don't influence too much the projected caustics, but I have to test with complex scenes.
B.Y.O.B. wrote: Tue Apr 10, 2018 9:36 am Usually the rule with render settings is: if you don't know what it is or does, don't touch it and use the default value.
Usually my rule with render settings is: touch all settings and ask to people that made the tool, to understand where a setting could be help you in certain situations. Sorry for this different vision but this rule work for me from 25 year ago when I start to work for 3DCG.
B.Y.O.B. wrote: Tue Apr 10, 2018 9:36 am If you want to know the whole story you should read an article about the Metropolis sampler.
I had read before. I understand generally what are the features of MLT and its priciple. I can't understand the matemathics, because it isn't my field.

Thanks for your hard work
User avatar
B.Y.O.B.
Developer
Developer
Posts: 4146
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: metropolis image mutation rate

Post by B.Y.O.B. »

Well there are two modes of usage: Experimenting and Production.
My answer was for production usage while you are probably in experimenting mode, so your view is perfectly valid :)
Post Reply