[kirkt]

My bad - i thought you were trying to see which renderer was most realistic. If they are all solving the same mathematical characterization of the physics, then they should all more or less produce the same mathematical solution, within the limits of the stochastic nature of some of the renderers' solutions. However, i thought the idea was to see if any of the renderers could reproduce reality. To do this I would think you would want to compare to a real reference.

I would argue that you can't necessarily assume that the math reproduces reality, even in highly idealized conditions -there are tacit assumptions in the math and its implemetation that make the solutions possible. Some solutions are pretty close to what we would think are real looking, maybe even experimentally verifiable under controlled conditions. Because the renderers all converge to the same solution only means that they are following, more or less, the same math model to get there. The question is, is the model an accurate representation of reality and is the implementation of that math acceptable both in appearance (e.g., splotchy cautics) and rendering time/computational overhead.

I'm not trying to knock your test, i think it is really a cool experiment. Because a lot of the output produced by these apps is a mix of quantitative and qualitative content and intent, there are always trade-offs of speed and realism. Thus, I am particularly interested to learn the outcome of your experiment.

[kirkt]

I posted above before reading your coin post. It's not that i don;t think the renderers can or can;t reproduce the nautral phenomena that are inherent in the model and the lighting. If a serrated edge on a coin focuses light it will create a caustic. Nature is always amazing. To think that we can completely characterize it is hubris to say the least - we do a pretty good job at coming close sometimes though. The best part is, that although nature sometimes seems to exhibit systematic behavior, there is room for interpretation, especially in areas like visual art. Light is a particularly cool concept.

A model of nature is only as good as its ability to reproduce reliably definable natural phenomenon. Even extrapolating beyond a specific validated model is an adventure.

Enough of my junk. Keep up the good work!

[Thomas An.]

My bad - i thought you were trying to see which renderer was most realistic

I would argue that you can't necessarily assume that the math reproduces reality, even in highly idealized conditions -there are tacit assumptions in the math and its implemetation that make the solutions possible. Some solutions are pretty close to what we would think are real looking, maybe even experimentally verifiable under controlled conditions

Thanks kirk, I don't mind a good conversation

About realism: As a first step we should see if we can get past some basics (such as the presence of caustic in the first place, as well as definition of caustic). The rest of the realism would most likely depend on each renderer's implementation of materials. For examlple in Maxwell the materials are based (at least partially) in spectral data, so the golds and silvers are pretty close to a physical speciment of the lab that produced the data. Other engines allow the users to define their own materials (and this inctroduces the human error in the picture). Overal, in this particular challenge the materials are simple metals (known metals) so it might not be to difficult for the users to come close.

In the photo bellow the placements of the light source is close but not indentical to the model (this challenge is not meant as a scientific experiment), but the elements of the caustic are discernible. The outer surface of ring in the photo is heavily scratched and the reflected caustic mirrors all those zillion microscratches. On the other hand the inner side of the ring has a more smooth finish and the caustic is more smooth ... as logic dictates, had the outer surface been smooth (mirror finish as in the model) the caustic would also be ultra smooth as in the Vray and Maxwell images)


I think the renders are using very good analytical models internally. Any error would most likely manifest itself as a "rounding error". So basically nstead of a caustic being 0.01mm to the left it is 0.01mm to the right. However these minute deltas would not produce a gross distortion of reality ... not in a simple image like this (since there is not much cumulation of error; as if it might in a mirror-in-mirror scenario). Also, lets not forget that the lenses of the Mauna Kea observatory (any high precision instrument) were once merely an analytical model in someone's computer, but their focal points were precicely predetermined within tight tolerances and when the physical components were manufactured the correctness of their mathematical foundation was confirmed.

Also, having said that ... I am fairly new to Vray, but as far as Maxwell is concerned, its correctness in terms of refraction indices and caustic callibration has been confirmed (mostly for the beta1.2.2a) in a number of more precise photo-vs-render experiements.

[hdesbois]

Interesting ...
The caustic is there, but very sparse... and the caustic on the inside of the ring seems to have a slightly different geometry than expected.

Right, but I think it's possible to get a better image. I'll post another one when it's cooked. Thanks for the challenge, it made me go deeper into lw renderer than I've been before.
HD

[LarsSon]

Interesting ...
The caustic is there, but very sparse... and the caustic on the inside of the ring seems to have a slightly different geometry than expected.

Right, but I think it's possible to get a better image. I'll post another one when it's cooked. Thanks for the challenge, it made me go deeper into lw renderer than I've been before.
HD

It seems that we can't get same kind of shape to the caustics.
Photon based renderers will produce only one bounce and there
will be only one shape per light inside the ring. Path tracers will produce
second, reflected shape to the opposite side too. So no matter if i
let Finalrender work 100h, there won't be secondary caustics anyway.

How can this be like that? In FR i can even render volumetric caustics, but
can't have secondary at all. How stupid!


-LarsSon

[big K]

not sure if this meets your new criteria, but this is my latest output out of finalRender st2 for cinema.
RT 4h35min on my PIV 3,2GHz (the caustics are still from point lights - as area doesn´t work at the moment)



i think it looks quite good, but it was nice to have a little reference like the maxwell rendering to see if all works out o.k.

the whole challenge is a lot of fun (and i also have lot about my render engines while working on this.) thanks thomas !

cheers
michael

[abgrafx3d]

Here's my first attempt:

RC5 - 6hr51min, SL 17.44, Dual 3.6gHz Xeon



Gonna make an attempt with Renderman for Maya next.

EDIT: Found out that Renderman for Maya does not support caustics in the current release, so I won't make that attempt. Maybe I'll give mental ray a shot.

[GM5]

Any idea what this dark line is? It is showing in the Maxwell renders and is not in the reference photo.



-Greg

[big K]

reference foto ? all i have seen are renderings so far
or did i miss something

[ajlynn]

It seems that we can't get same kind of shape to the caustics.
Photon based renderers will produce only one bounce and there
will be only one shape per light inside the ring. Path tracers will produce
second, reflected shape to the opposite side too. So no matter if i
let Finalrender work 100h, there won't be secondary caustics anyway.

How can this be like that? In FR i can even render volumetric caustics, but
can't have secondary at all. How stupid!


-LarsSon

**sticks head in room**

Not true. FR2 does this, so does mental ray and probably Vray. Here's one in FR2 (it's not very good, but it has those caustics and took about 40 min on a P4-2.4)



GTG.

[GM5]

reference foto ? all i have seen are renderings so far
or did i miss something

The one in Thomas An's post on this page...

-Greg

[big K]

oh this one.

i would say it is a different geometry, therefore different caustics.

michael

[LarsSon]

It seems that we can't get same kind of shape to the caustics.
Photon based renderers will produce only one bounce and there
will be only one shape per light inside the ring. Path tracers will produce
second, reflected shape to the opposite side too. So no matter if i
let Finalrender work 100h, there won't be secondary caustics anyway.

How can this be like that? In FR i can even render volumetric caustics, but
can't have secondary at all. How stupid!


-LarsSon

**sticks head in room**

Not true. FR2 does this, so does mental ray and probably Vray. Here's one in FR2 (it's not very good, but it has those caustics and took about 40 min on a P4-2.4)



GTG.

Wow! There you have it. I have tried to do the same, but it
just won't show up. Well, there is differencies with FR2 and FR1 for Max,
but could that just be the reason?
How many photons are you shooting in that image?

-LarsSon

[ajlynn]

Photons in the millions, FR2's metal shaders, FR Object tags on the reflective parts with caustic energy at 100. Don't know how much of that translates to FR1.

[glebe digital]

I'm so surprised how poorly Lightwave has performed in these tests, I was certainly expecting better.


Ok maybe it's not terrible, but it's not going to 'make the grade' on these test conditions and those Mental/Vray images just leave it for dust.........set-up time is hideous [full day of fixing/tweaking] and the caustics are not convincing in the least.


This is the final Maxwell treatment but is still cooking as it needs to clean up on the inside faces of the two rings........hopefully by tomorrow morning it'll be complete.
Thanks for the test Thomas, it's been interesting. Will this last M~R image pass the criteria ok?