BDPT Translucent Glass Luminaries

[anchovy]

Yesterday I stumbled upon a couple cool images of some blown glass luminaries. The cool thing is that its a pair of images one lit with a studio setup and the second dark with only the luminaries providing lighting. Im immediately thinking this is a good test, can I using the the BDPT and a single material per lamp come close to the two reference images?

Here are the two photos.



And here are the two renders.



Not too shabby, for a relatively quick poke at the look dev. that being said there are a couple of interesting issues. It was near impossible to find a happy medium for the lit externally and lit internally glass materials. This is especially true for the large hue difference in the peach colored glass. At some point I stopped playing with it and landed on the color you see. The glass is heavily glossy in the refraction and i'm using a sphere light to illuminate the material from within. The second problem is that without using an inverted hue of the base color in the refraction swatch the glass goes dingy.

Here is what the materials look like by desaturating the refraction color for each material but leaving it at the same value. note that the don't look as bright and saturated. the tradeoff is you can get a bit of an opposing color cast, the violet in the yellow for example.

[anchovy]

Heres the dark version.



Here are the chromatic settings for the refraction for the various materials.



This begs the question, To not have the transmitted color be the inverse of what I would expect i'm playing fast and loose with the refraction/absorption. Is there a better/more correct way to approach this. Ist the same type of problem you get with SSS. I gather its physically correct which I like conceptually, but the results are not wonderful and for most users not so intuitive.

i just dont see much in the way of examples in real life where orange juice looks green. Id love to hear what anyone thinks.

Thanks!

-Michael

P.S. it may be a bug and if so ill post it but the sphere light creates what appears to be a hemispherical delineation on where it emits, almost lit it isn't sampling across the whole sphere. Ive tried it with a mesh light and you don't see the same artifact.

[anchovy]

A small update. Not so bad, for the look.



-Michael

[andybot_cg]

Neat, looks nice! But geez, did it really run for 59 hours!! I recall there was a thread a little while back that someone was trying to render an orange juice material with a buddha statuette. There were issues there with the opposite color cast, and I think using fog color only resolved the issue. Did you try this test with fog color only and no refract color?

[anchovy]

Yep that was me. And you can see above the no refraction color is a lot less saturated hence the trying to add in a bit of color to the refraction. As to the time, yeah.... I'm not trying to mack this production worthy, its really more of a pressure test for the BDPT. I want to know where the edge of the envelope resides. Its hard to stomach but it always takes twice the number of samples to reduce the noise in half with monte carlo methods like path tracing. Hence if you want to have the image fully converge you could be sitting there awhile.

But back to the point in hand, I'm looking into why it is that the way translucency/sss is set up and is physically correct. And I have no doubt that Vlado has it right, but that in the real world we just don't see this phenomenon very often. It has everything to do with the colors that are absorbed vs. the transmitted remainder. Funny thing is we just don't see green in our OJ.

-Michael

[andybot_cg]

could you add back the color with higher fog multiplier/ increased fog color saturation?

I've used bdpt with other renderers, and it's just not very satisfying to use for anything production because it will never be clean. The noise looks more like dust and speckles than the nice film grain quality of dmc.
(and you might get a bitchslap from c nichols for using it )

[anchovy]

Hehe, its just a matter of time, till we are all using it. Some day we will look back at any of the sparse sampling techniques and wonder how we possible got anything done. A good example of this is they are just now switching over to using a more physically based lighting model at pixar and they have found out that lighting is now taking half the time it used to...

-Michael

[andybot_cg]

Sure - at pixar levels they can throw the render farms at the problem, but I don't know about the small fish...

So what do you think of the noise with BDPT? I personally don't like it at all.

[vlado]

So what do you think of the noise with BDPT? I personally don't like it at all.

It will get better

Best regards,
Vlado

[anchovy]

The idea is it converges to no noise eventually And the point I was hoping to make with the Pixar story was more of, sometime a stack of band-aids is less cost efficient than just brute forcing it. We currently us a lot of "Tricks" to get some of the subtle goodness that simpler methods get for free/just time.

It will be fun to see where it goes.

-Michael

[Morbid Angel]

It will get better

Just let it run for another 500 hours

[anchovy]

Just let it run for another 500 hours

Zing!

-Michael

[sherridge101]

Hey all,

is it possible to resume the render passes at a later point?
So you could start any compositing work on a lower passes version and then resume the render in the background.

cheers,

Steve

[sherridge101]

And are the caustics switches in the VRayBPTracer for direct and indirect caustics?

cheers.