Mixed Renderer Sampler

[Meelis]

I find random sampler works better for some darker sss volumes and MLT works better for lighter transparent sss volumes.

So idea is that i mix maybe 70% MLT renderer sampler with 30% random sampler.
If i can set LuxRender to render 42 min with MLT and 18 min with random and automatically switch between.

Random gets some values and i set MLT as greater importance .
MLT film would be main and LuxRender looks on MLT layer if there is much darker area compared to random layer and then balances this darker MLT pixels.

[Vutshi]

Isn't it possible to just tune MLT parameters so that it works more like random sampler?

[SATtva]

Right, just set large mutation probability to something like 0.8.

[Meelis]

Oo okay, didn't know that.

[jeanphi]

Hi,

SATtva is right, what is called large mutations are actually uncorrelated mutations, they are fully random over the entire sampling space.

Jeanphi

[Meelis]

Hi,

SATtva is right, what is called large mutations are actually uncorrelated mutations, they are fully random over the entire sampling space.

Jeanphi

Hi, thx for answer. Can someone make explanatory diagram in wiki how the 0 - 1 mutations correspond to MLT vs random sampling balance.
Maybe it's linear like 0.1 is 90% MLT and 10% random.

[jeanphi]

Hi,

Both mutations are part of the MLT algorithm.
The small mutations try to explore nearby space in search of a maximum, it proves effective because with a small displacement you won't go over the maximum in a pit. However you can get stuck at that maximum and completely miss other maxima.
That's why once in a while a completely random sample is tested. The tuning knob is really the proportion a large (random) mutations vs small mutations.

Using the standard Blender interface, the MLT strength parameter refers to the proportion of small mutations while in advanced mode, you tweak the large mutation proportion. It's completely linear in either case.

Keep in mind that if a mutation (even a large one) is deemed inappropriate it is going to be rejected, that's why the default value of 40% large mutations is used even if it might appear quite high at first.

Jeanphi

[Meelis]

Thx jeanphi, for detailed answer.

[J the Ninja]

Btw, in 2.6 we don't use the "strength" param, you set large mutation probability directly either way. The advanced option just shows/hides maxrejects and use variance.

[Meelis]

Btw, in 2.6 we don't use the "strength" param, you set large mutation probability directly either way. The advanced option just shows/hides maxrejects and use variance.

Look at that, very very nice
Lower is newer LuxBlend