Brute Force Caustics

[Macker]

Could this be done? Kind of like how brute force GI works... A subdivs value, and a multiplier?

[3LP]

Caustics are often asked, including me
From my understanding, and I hope I'm wrong, CG still don't know who to implement this efficiently.

Stan

[LudvikKoutny]

You can have them. Just set both primary and secondary GI to brute force, make sure both reflective and refractive caustics are enabled, and disable max. ray intensity. They will take ages to resolve though

[Macker]

You can have them. Just set both primary and secondary GI to brute force, make sure both reflective and refractive caustics are enabled, and disable max. ray intensity. They will take ages to resolve though

Just did a quick test, didn't work. Those are GI caustics, not "regular" (direct lighting) caustics, no?

[Neilg]

Isnt this how they work if you turn on direct visualisation? You've got subdivions in the light and an overall scene multiplier.
I've been doing caustics like this for years, leave 'dont render final image on' and you can render it as a separate pass with only the lights and objects you want caustics on in the file.

[vlado]

Isnt this how they work if you turn on direct visualisation? You've got subdivions in the light and an overall scene multiplier.

Yes, that's what I was going to say. For the moment that's the only alternative. Although I'd also set the max. density to something other than 0.0 so as to keep memory usage under control.

Best regards,
Vlado

[Macker]

I see, you learn something new every day. I shall give this a go tomorrow. This is still photon mapped though, yes?

I just find it odd that every single aspect of rendering (not just with vray) has moved away from photon mapping, yet it still remains the go-to solution for caustics. Why is this?

[vlado]

I see, you learn something new every day. I shall give this a go tomorrow. This is still photon mapped though, yes?

Sort of.

I just find it odd that every single aspect of rendering (not just with vray) has moved away from photon mapping, yet it still remains the go-to solution for caustics. Why is this?

Nobody has found a better way to do it so far... there are special algorithms for some specific cases, but not for the general case. Even engines like iray and Maxwell trace photons from lights in order to capture caustics effects.

Best regards,
Vlado

[Macker]

Why does photon mapping suit this so well and not other methods?

[vlado]

Why does photon mapping suit this so well and not other methods?

Because finding small light sources through multiple reflections/refractions with path tracing is difficult and happens with very low probability.

Best regards,
Vlado

[Macker]

Sorry to drag this back up, but would it not be possible to implement some kind of hybrid-biased/unbiased solution? Perhaps something that initially seeks out blurry/rough caustics (biased) and refines the result over time, gradually becoming more unbiased?

Does metropolis light transport not do a similar thing, but with the entire lighting solution?

I'd love to be able to have caustics permanently on, so that windows\large expanses of curtain walling reflect sunlight back down onto the road without using tens of gigabytes of RAM.

[saudade]

everything is possible with programming.

[peterguthrie]

everything is possible with programming.

apart from caustics

[Dariusz Makowski (Dadal)]

Just gotta wait for deepe learning to do caustics I guess... :- )

[Art48]

Just gotta wait for deepe learning to do caustics I guess... :- )

What would actually be kind of funny if they implemented it the other way around. Don't use deep learning to find an algorithm to do efficient caustics but use deep learning for the "AI" to look at your picture, decide where there would be caustics, then creates them out of its "mind" as it sees fit.
We mostly just need correct looking caustics, not neccassarily physically correct ones :P