LuxCoreRender Forums

Yes but that is a Metropolis problem, not a PSR one.

That's what I've been thinking about the last few days.

Doesn't PSR mollification have to happen during MLT small mutation?
Large mutation phaseses dont know about each other, so an old phase wont help the next to find a path...

So if I currently set the pathspaceregularization.speed to 1... and let's say scale to ~1(or whatever value). Won't I end up with the same result as if mollification was removed?

And if that's the case, then the only purpose of PSR is to make sure sampled caustics are never smaller than set value... And if that is true, wouldn't some sort of progressive photon mapping make more sense than using PSR+metropolis?

Dade wrote: Thu Sep 19, 2019 10:16 am...both PSR (on angles) and VM (on space) are just a theoretical exercise to use a trick: to replace a perfect specular glass with a slightly glossy glass.

Using the cap for the very beginning and removing any reduction process is exactly what I have done and works better:

Now I'm confused, without mollification, what is left from PSR?
What is the difference to path+metro+very slight glossiness?

Edit: After thinking about it: PSR would be light tracing till specular hit, only last connection to camera done with (extra) Metropolis exploring small angles.
Path + Metro + glossiness would mean Metropolis sampling the whole path from camera to light which is probably more difficult.

Suggestion: The angular difference between two neighbour pixels (angle/field of view) could be used as cap granting pixel-sharp SDS-caustics

provisory wrote: Thu Sep 19, 2019 11:00 am That's what I've been thinking about the last few days.

Doesn't PSR mollification have to happen during MLT small mutation?
Large mutation phaseses dont know about each other, so an old phase wont help the next to find a path...

I mean, the reduction of the mollification bandwidth has to happen between MLT small mutations not between large mutations.
And on large mutation the bandwith should be reseted.

This way the MLT easily finds the SDS paths, than explores them with increasing accuracy thanks to the decreasing bandwith.

provisory wrote: Thu Sep 19, 2019 12:13 pm

provisory wrote: Thu Sep 19, 2019 11:00 am That's what I've been thinking about the last few days.

Doesn't PSR mollification have to happen during MLT small mutation?
Large mutation phaseses dont know about each other, so an old phase wont help the next to find a path...

I mean, the reduction of the mollification bandwidth has to happen between MLT small mutations not between large mutations.

It is exactly the contrary according the paper (and it makes sense).

Dade wrote: Thu Sep 19, 2019 2:15 pm It is exactly the contrary according the paper (and it makes sense).

You mean the paper or my idea makes sense?

BTW refined my above description a bit.

lacilaci wrote: Thu Sep 19, 2019 11:18 am So if I currently set the pathspaceregularization.speed to 1... and let's say scale to ~1(or whatever value). Won't I end up with the same result as if mollification was removed?

Yes, exactly (indeed, the .speed parameter has been removed in the very last sources).

lacilaci wrote: Thu Sep 19, 2019 11:18 am And if that's the case, then the only purpose of PSR is to make sure sampled caustics are never smaller than set value... And if that is true, wouldn't some sort of progressive photon mapping make more sense than using PSR+metropolis?

They are all pretty much the same: in PPM, you have lookup radius reduction and there is the same problem, you can not reduce the radius "ad infinitum".

provisory wrote: Thu Sep 19, 2019 2:18 pm

Dade wrote: Thu Sep 19, 2019 2:15 pm It is exactly the contrary according the paper (and it makes sense).

You mean the paper or my idea makes sense?

The paper

Dade wrote: Thu Sep 19, 2019 2:24 pm

provisory wrote: Thu Sep 19, 2019 2:18 pm

Dade wrote: Thu Sep 19, 2019 2:15 pm It is exactly the contrary according the paper (and it makes sense).

You mean the paper or my idea makes sense?

The paper

To further elaborate: the paper makes sense in the context if this infinite angle reduction ... in practice it may be different, I can try that.