Which render mode is closest to reality?

[zeitmeister]

Hi,

I just had a look into the actual Indigo manual... and I wondered which ray tracing method is the one which is closest to reality.
As I understood unbiased renderers (and I always presumed that Indigo were such), the light rays are being shot from the light sources on the objects and then into the camera.
regarding the schemes in the manual, I can't find such a mode working like that.

Plus:
In Cindigo, I can choose between pathtracing, bi-directional pathtracing, MLT and bi-directional MLT.
So which ones refer exactly to the modes described in the Indigo manual?

So I wondered which render mode gives the most realistic results? I know I could test and test and test, but I'd like to choose the one which works most realistic and then work with that.

Thank you in advance!

[Zom-B]

Afaik all of them should render the same result after Time X. The reason for different Methods are the different Strengths of the algos.
PT is great for direct light, MLT is faster with caustics etc., and BiDir is a general tech that starts ray from light & from cam to enhance performance.

We all wait for the great Hybrid Method to come back to Indigo (announced by Ono some time ago!), so Indigo will combine MLT+PT in one rendering progress and choose the best algo for "materials/parts/whatever" of the scene

A quite nice article can be found on the german wikipedia side: http://de.wikipedia.org/wiki/Raytracing

[galinette]

Yep : all algos converge to the same solution.

Depending on the scene, some converge faster, some slower. Also, for a noisy image (short render time), some algos generate somewhat esthetic noise, some do ugly noise. In my scenes with lot of glass, MLT generally converge faster, but the first images are uglier, because instead of making smooth uniform noise, it makes large colored strokes.

MLT was initially designed for scenes with highly indirect lighting, the case where it shines most is a room lit only through a slightly opened door. In these scenes, without MLT, most of your calculated samples end up with no contribution to the image.

Also, I have found that for complex glass pieces, with multiple internal reflections, MLT is also very effective (One example is dispersion_test.igs : without MLT, the images looks OK at the beginning, and some bright dots start to appear after some time. In fact, these bright dots would converge to a multireflection effect after a very very long time. MLT makes the renderer focusing its efforts on these reflections and spends more samples there, resulting in a much faster convergence).

I have made less tests of Bidi-NonBidi. But it appears that bidi helps when you have "concentrated" light sources, like bulbs and so on. When using envmaps with no very highly concentrated light spots, I have found bidi to slow down a lot the renderer with no benefit.

Etienne

[zeitmeister]

These are helpful infos, thank you!

What remains is the "confusing" naming of the modes between the manual and Cindigo.
fused?

[pixie]

These are helpful infos, thank you!

What remains is the "confusing" naming of the modes between the manual and Cindigo.
fused?

Could you elaborate on that?

[zeitmeister]

Oh, me and my English... what do you exactly mean with "elaborate"?

Here's a screenshot of the renderer modes in Cindigo:

modes_cindigo.gif (7.37 KiB) Viewed 5285 times

And in the official manual:

modes_manual_01.jpg (37.16 KiB) Viewed 5285 times

modes_manual_02.gif (20.88 KiB) Viewed 5285 times

[pixie]

Elaborate as in explaining it a bit further, just as you've done, thanks. Indeed it seem a bit dawning, I never thought it to be any difficult since I never checked the manual, and since only 4 modes are actually available I never expected it to be confusing, I was wrong... The QMC bit puzzles me a lot

[fused]

QMC is always used when MLT is not.

[pixie]

The reason of my confusion is having 4 images while lacking the representation of MLT bidir and having 3 representations of PT while having just two PT modes and 2 MLT ones.

[zeitmeister]

.... Exactly.

[Soup]

Pictures are illustrated from Ono's sketches

The pictures don't directly show the available modes. They show them separately. But it is confusing that it doesn't simply illustrate the modes that appear in the exporters.

Ok I whipped up a quick change, this reflects the options listed in your exporter.
Does that help? I want the manual to be as clear as possible

[CoolColJ]

Bidirectional MLT is probably the most robust of all the methods

It can find and render complex light paths better than the other ones

[dcm]

Bidirectional MLT is probably the most robust of all the methods

It can find and render complex light paths better than the other ones

word

[zeitmeister]

Alright, thank you!

[forgeflow]

Bidirectional MLT is probably the most robust of all the methods
It can find and render complex light paths better than the other ones

Except, it has a bit of bug regarding rendering of glass - it tends towards black edges, whereas rending with MLT alone does not produce this artifact. Sadly, rendering with MLT alone is also a lot slower than MLT BDir so you are still in the position of having to chose the right method to suit the scene.