LuxRender 1.0-dev Windows build of 22-05-2012

[A-man]

Such as?

You sometimes get these overbright spots/areas that never fully converge no matter how long you render it out. Which is something ERPT was supposed to counteract

In scenes with complex refracted caustics highlights, it's pretty much luck if you get any of them to show. You will get some, but it will be random from render to render, and you will never get all of them.
That's the thing, too much luck involved to be called a robust rendering solution.

MLT is after all, a random walk rendering style. I wish there was something better now that was not left to chance.
MLT is damn old in computer years!

Wait isn't MLT usually used with bidir? Because it seems to me that caustics converge very well and consistantly with that. AFAIK, MLT is not just random, it is a very smart algorithm.

Exhibit A: viewtopic.php?f=15&t=8192&p=82001
Exhibit B: download/file.php?id=10208
Exhibit C: viewtopic.php?t=7806&p=77331#p77331 (note the caustics visible through the scattering medium on the bottom of the box)

It seems you are just bashing MLT for no reason

[CoolColJ]

Such as?

You sometimes get these overbright spots/areas that never fully converge no matter how long you render it out. Which is something ERPT was supposed to counteract

In scenes with complex refracted caustics highlights, it's pretty much luck if you get any of them to show. You will get some, but it will be random from render to render, and you will never get all of them.
That's the thing, too much luck involved to be called a robust rendering solution.

MLT is after all, a random walk rendering style. I wish there was something better now that was not left to chance.
MLT is damn old in computer years!

Wait isn't MLT usually used with bidir? Because it seems to me that caustics converge very well and consistantly with that. AFAIK, MLT is not just random, it is a very smart algorithm.

Exhibit A: viewtopic.php?f=15&t=8192&p=82001
Exhibit B: download/file.php?id=10208
Exhibit C: viewtopic.php?t=7806&p=77331#p77331 (note the caustics visible through the scattering medium on the bottom of the box)

It seems you are just bashing MLT for no reason

Those are easy caustics to render
When you can directly view the surface the caustic is projected onto, it's not too hard.

Try rendering caustics that are on a bump mapped surface that is sitting under water/glass and your viewing it from above the water/glass. Much tougher, especially if it's lit by the sun.
And not just render them in a general way, but every single caustic detail to be exactly the same everytime you render it

and try this - multiple glass objects bouncing and refracting caustics to each other when lit by the sun.
2 different renders - 2 different results...

download/file.php?id=16420&mode=view
download/file.php?id=16407&mode=view

[SATtva]

All rendering settings were identical (bidir 32/32 bounce depth, light strategy "one", mlt lm.prob 0.2). Hybrid mode has shown C/s value more twice higher than CPU mode, however the image looks less converged and a lot darker.

Dade, i can send you the scene if you'd like to play with it.

The additional noise is normal and comes from not having MIS. There is a partial implementation of MIS in hybrid BiDir I'm working on and can be enabled with an integrator flag. To have an idea of the difference just try to render the gold Luxball with and without MIS ... the difference is astonishing.

Wait, MIS must be enabled explicitly? I thought it's always on like with CPU Bidir, otherwise i think we're missing the corresponding toggle in LuxBlend (or it's me missing something).

The difference in light intensity shouldn't be there, can you isolate the problem (i.e. just wall+light source scene) and post the lxs/lxm ?

Sure will do.

[SATtva]

Try rendering caustics that are on a bump mapped surface that is sitting under water/glass and your viewing it from above the water/glass. Much tougher, especially if it's lit by the sun.

These are limitations of path tracing algorithms, not the sampler, and exactly the perfect applications for SPPM or BidirVM.

[CoolColJ]

Try rendering caustics that are on a bump mapped surface that is sitting under water/glass and your viewing it from above the water/glass. Much tougher, especially if it's lit by the sun.

These are limitations of path tracing algorithms, not the sampler, and exactly the perfect applications for SPPM or BidirVM.

Hopfully BPT+Vm can do it, because even SPPM is not so good at rendering diffuse surfaces unless they sit under glass/water
And definitely does not like glossy/phong surfaces if you have a lot of them.
SPPM also does not render bounced and refracted reflections/specular highlights between glass objects that you can see hints of in MLT, but again as I mentioned this is rather pot luck and random with MLT as well
Just has a hard time finding all of these features

[zeealpal]

But for what reasonable reason could you possibly need to render these caustics for? Apart from for the sake of.it?

[CoolColJ]

But for what reasonable reason could you possibly need to render these caustics for? Apart from for the sake of.it?

accuracy! And the fact it looks ugly with incomplete blotchy caustics and hotspots...

[zeealpal]

You could always Photoshop/Gimp that area, adjust FF rejection? Or do a border render of that part with the same linear tonemaping settings?

But then again, I like to experiment alot too

[Lord Crc]

When Lux crashes, it just gives the standard Windows close program, or ask MS for help (which is of no use, obviously) dialog. Can the Windows builds be modified to write a minidump when they crash?

I have no idea how to do that, I'll have to look into it. They should have debug info in them, so if you have Visual Studio you could attach it to the lux process before rendering, that should at least give you a stack trace.

[cwichura]

I have no idea how to do that, I'll have to look into it. They should have debug info in them, so if you have Visual Studio you could attach it to the lux process before rendering, that should at least give you a stack trace.

I believe the folks here, for our Windows software, have implemented the unhandled exception handler in Windows (Google the SetUnhandledExceptionFilter() API call) and write the dump information in that. Windows calls that for any uncaught exception, instead of doing its usual dialog. You can save out whatever you need/want during that, and also throw up your own dialog saying 'i've just written out file xyzzy, please submit that to blah to help us diagnose the problem'.