Caustics with PT-MLT and BI-MLT

[Headroom]

I created this little tank thing to help another user with how to make his tank look more realistic.

I rendered a version using Pathtracing with MLT which rendered for about 5 hours. I started with Biderectional with MLT but the results are very well....discolored, perhaps a bit too strong dispersive effect ?

Anyway, I rendered the BI-MLT version through the night just to see if the "colorized" effect would dissappear when rendering longer. The BI-MLT version rendered for about 12 hours and while the effect is much reduced it's still quite visible.

What would explain the difference ?

[galinette]

To me, it looks like:
- The BI-MLT version is characteristic of dispersive caustics rendered not long enough with MLT
- The PT-MLT has absolutely no colored noise but seems not fully rendered yet (caustics are noisy and hairy). That sounds like dispersion is disabled... Quite strange

Etienne

[fused]

Hey Headroom,

which version of Indigo are you using?

[Headroom]

I have been too lazy to test 2.6.1 so I am still using 2.4.13.

[Meelis]

Bidir is just VERY slow.
100 h bidir outdoor render is often no cleaner than 2h path tracing render with specular materials dominating at least. I soo the big speed difference with high poly 10M tri Lucy model.

[Headroom]

Strange indeed.

I looked at the material with the material editor and it does not have cauchy_b enabled

I am not sure I perfectly explained how I used displacement. I used only Blender's internal displacement modifier which was based on a cloud texture. While this creates a heavy mesh when exported it provides nice interactive feedback directly in Blender.

I tried today to create the same effect with Indigo features/functions and ran into problems.

I baked the cloud map from Blender into an image.
Then I modified the material in the material editor to use that image as a displacement map with 40mm displacement value. In Blendigo I used 2 levels of subdivision and subdivision smothing.

While the water surface (Vertex group in blender for surface only and UV unwrapped) was nicely displaced it created a nasty black rim where it joined with the rest of the mesh. Also the rest of the mesh while not affected by the image map was displaced "linearly" (if that makes sense) so that now a good part of it was protruding through the tanks surface. Reducing the mount of displacement from 40mm to 20mm almost fixed it. Just almost.

And then I really had to get some work done

[StompinTom]

As far as I know, subdivision is a bit borked right now in Indigo. To be fixed in the next version, though.

[CTZn]

Make sure that the mesh to be displaced has all edges soft, that's a big helper.

Displacement is currently borked in a sense wich makes it hard to keep under control if you didn't get all the workarounds it takes (hard edges will explode the displaced mesh, UV borders and face rings bording them must be hidden if subdivided at render time, essentially).

Yes, Ono demonstrated publically large improvements to come

[Zom-B]

Displacement is currently borked

Just to clear this: Not displacement is broken, but Indigos subdiv!
You can end up with great displacement if doing subdiv in your 3D app and export a highpoly mesh!
(But RAM usage and startup time are higher)

[Headroom]

This one was rendered with Indigo 2.6.1, Bidir-MLT, Architectural Glass Acceleration.

40+ hours. iMac i7, 2.93GHz.
I don't know exactly what he SPP were but it was over 12000.

"Clear Natural Water" Material w/out cauchy_b from MatDB.
I needed to correct the precedence setting for the glass, which is the reason the caustic pattern is slightly different from the previously posted images.

This is completely subdivided in Blender and exported as a high-poly mesh.

[Borgleader]

Soooooo pretttyyyyyy

[Meelis]

Then it comes from phong material.
I do get this color elephants from phong exp 5000 carpet.

3 super samples

Phong -VS- Diffuse

[CTZn]

Not phong alone Meelis, but the fact that it is of an high exponent and blended by an amount with the null material while using bidir, not to mention the association with the complex glass material and mesh. That's a lot of cumulated factors that go well beyond the topic of caustics.

Monochromatic noise in PT was recently introduced with reason by lycium, while MLT will (probably) always be colored hence slower to converge, but also safer in such a complex scene.

Each mode has its strong points and drawbacks altogether, it's up to the user to keep the balance

[Headroom]

Monochromatic noise in PT was recently introduced with reason by lycium,
)

You're right. Now that you mention it I remember

Given the simplicity of the scene/object I believe this makes a perfect example to demonstrate the effect of Luminance vs. color noise.

Should be filed under "Indigo - The missing Manuals"

[CoolColJ]

I'm an expert when it comes to rendering water

That's my water material. Anyway it seems with Indigo, if the light hits the water at an acute angle it's very slow. The closer you can get the light source to 90 degrees with the water surface the faster it renders, as far as caustics go.

And if the surface the water is sitting on has bump mapping of displacement, that will also increase render times of caustics