Liquids Test Thread

[WytRaven]

To be honest, although the env map version does look nicer the difference in rendering time just isnt worth it for a test scene. We need speed over sex appeal

I also notice that it converges significatly faster without the env map. compare 10Kspp (no env) vs 63Kspp (with env)...big difference in the liquid convergence if you ask me...


I'll be interested to see how an env version with exit portals goes...

[PureSpider]

I also notice that it converges significatly faster without the env map. compare 10Kspp (no env) vs 63Kspp (with env)...big difference in the liquid convergence if you ask me...

Noticed that, too.

I'll be interested to see how an env version with exit portals goes...

The env version with exit portals is pretty much the same as without them as the exit portals would be at the same position as the scene boundaries are anyway... or am I wrong here?

So let's stick with the no env version for the sake of rendering speed.

€dit: Optimized the igs, removed unused materials and so on.
€dit 2: Changed the background position a bit so the grid texture fits to the pot's body.

[neo0.]

I don't know why this never occurred to me before, but in real life orange juice never turns black unless it's geared towards a certain scale. You can pour it into a small cup or a big one and it still looks very much like orange juice.

[Borgleader]

I don't know why this never occurred to me before, but in real life orange juice never turns black unless it's geared towards a certain scale. You can pour it into a small cup or a big one and it still looks very much like orange juice.

.... What? Are you referring to the bidir bug or whatnot?

[neo0.]

I don't know why this never occurred to me before, but in real life orange juice never turns black unless it's geared towards a certain scale. You can pour it into a small cup or a big one and it still looks very much like orange juice.

.... What? Are you referring to the bidir bug or whatnot?

Nope. My OJ turned black when the scale was too small. That was the problem that a few people discussed with me on the previous page.

[Borgleader]

I don't know why this never occurred to me before, but in real life orange juice never turns black unless it's geared towards a certain scale. You can pour it into a small cup or a big one and it still looks very much like orange juice.

.... What? Are you referring to the bidir bug or whatnot?

Nope. My OJ turned black when the scale was too small. That was the problem that a few people discussed with me on the previous page.

That thingg about small faces turning black?

[neo0.]

Yep. The whole discussion we had about scale. In real life, oj still looks like oj whether it's in a 2 inch of 5 inch glass.

EDIT : Bidrectional MLT = 31,000 samples/sec and plain MLT yields around 60,000 samples/sec. Around twice as fast.

[Borgleader]

Yep. The whole discussion we had about scale. In real life, oj still looks like oj whether it's in a 2 inch of 5 inch glass.

Yes, and that's a bug right? And isn't it related to bidir? like if you use the same scale and turn off bidir, do you still get it?

[neo0.]

Both are at around 30 spp
]Bidirectional MLT



MLT


Notice the slower convergence on the first one. Not only do the samples take longer, but it will also need more of them to obtain the same result as MLT.

[PureSpider]

Yeah yeah finally learn how one does liquids in a glass right, okay?

[Whaat]

The two renders show some interesting differences between Path Tracing and MLT (bidirectional turned off for both renders). The PT result is the best quality with almost none of the blackness around the rim of the glass, the meniscus of the juice, or the base of the glass. I wonder if this a bug with MLT or if the settings for MLT just need to be tweaked. I only had Max Num Consec Rej set at 500 for the MLT render. I will try another render with it set higher.

If you recall from my other posts, when bidirectional is turned on, there is significant black artifacts for BOTH PT and MLT methods. However, I am not posting examples of that here.

The PT version rendered much longer than the MLT one did. However, I think they both converged at a similar rate. I just stopped the MLT one before all the noise cleared.

Tests are with Indigo 2.0.12

[Borgleader]

Nice nice Could you post your scene? I could also let it render overnight, I have 4Gb of RAM & a 2.4Ghz quad core so I think I should be able to handle a pretty high setting.

[OnoSendai]

Hi Whaat,
just remember that any non-infinite M.N.C.R is actually biased. And 500 is a rather small value - I don't think you're giving the MLT algorithm enough chances to sample the tricky areas of the path space.
Try cranking up the MNCR to a million or so

[Whaat]

Hi Whaat,
just remember that any non-infinite M.N.C.R is actually biased. And 500 is a rather small value - I don't think you're giving the MLT algorithm enough chances to sample the tricky areas of the path space.
Try cranking up the MNCR to a million or so

Yes, you're right! This is the ticket! I am rendering a new one with MNCR at 2000 and the result is looking much better!

[matsta]

thanks for wasting like a million megs of my hard earned cap on the two ugliest images i think i've ever seen neo! ill drink to you this evening.

@ Pure, that last teapot was hot!
@ Whaat, biggest problem with OJ renders is that nobody wants to model pulp :< until someone does that they will never truely be accurate.

Loving this thread btw!