[polynurb]

following this thread i was eager to find out some of the optical workings behind the systems we use quite every day.

since maxwell is what it is, i expected it to be somewhat capable of tracing accurate results, questionable was more how well it will handle the transport of multiply reflected/refracted caustics within a given scene.

the first task would be to build a single lens optical system to make sure only first order caustics are used to light the scene (most real projectors use fresnel+lens triplets.. for that our computers are too slow..) and trying to keep spherical aberration as low as possible. (as long as we leave dispersion turned off there won't be any chromatic aberration )

in the tests i did it appeared that a planoconvex lens performs better in terms of spherical aberration than a biconvex lens,
so that is what you see being used to create the image above. (kudos to Brenda for assisting in the shot)

very simple 2 element design:


"DLP" chip is a sqaure mxi emitter plane

projected image:


lense is a single planoconvex element, spherical side facing outwards. (should be better choice that way around if distance to object is much bigger than distance to back focal distance)

enclosure (to aviod stray light) is made of an "absorber" material, nd 1 black black lambert; the box is closed.

there are a few relatively simple equations one can use to determent the correct distances of a certain configuration:

the thin lense equation

the "lensemaker's equation

both described here:

wikipedia is a good read but an outstanding description about this matter can be found here

> http://www.handprint.com/ASTRO/ae1.html#thicksingle (thanks to bruce for the hard work)

so the system needs to be focused in order to project an sharp image; that is a bit prblematic, as you will prbably know from setting up a beamer, that a minimal turn on the focus wheel has a big effect on image sharpness. So the setup has to be very accurate to re-project the texture accurately and it will only work on the target distance it is focused to.

it should be possible to construct a dynamic working system in a parametric software like grasshopper(rhino plugin), which will be the next thing i'll look into once time allows.

btw. i am not in any way an expert in this field.. but it is relatively easy to grasp the basic concept;
give it a try!




-daniel


ps. FIRE is not a good choice for testing such a setup
apart from that render times are absolutely acceptable.. what an engine..

[Mihai]

That is one of the most interesting images I've seen.....something about it, like a photo of an unreal scene. Nice work!

[eric nixon]

It will always be somewhat unreal as the caustic rays wont converge until sl30+.

What I mean is the emitter behind glass approach is a big no no... no?

apart from that render times are absolutely acceptable.. what an engine..

But the reflections of caustic light will take forever to develop..

[polynurb]

thanks mihai,

for such testing quickshots, i find it's often best to finish before you realize what you are actually doing

[polynurb]

What I mean is the emitter behind glass approach is a big no no... no?

now that was unavoidable in the test setup.. but technically you are right just as we still can't render day lit interoirs with real-glass windows.

But the reflections of caustic light will take forever to develop..

the image is sl25, and the only artifacts in terms of noise are on the mirror elements; these improved quite a bit between sl20 and 25. according to the calculator rendering this to sl32 would take 333h, whereas i rendered this one over night.
what still works very well however is the reflection off the mirror onto the wall (behind the charater) or the floor (stripes and dots) and i don't know any other way in MXW to achieve this..

in fact these mirrors already reflect the caustics they produce themselves in a decent way, so i find this quite usable. although i would not render something like that in a multilight setup but as a separate pass for comp.

what i noticed though is that the projected image itself (on the wall) keeps becoming sharper over time, which must be due to the pure amount of rays transmitting the lense to recreate the image..

-daniel

[eric nixon]

in fact these mirrors already reflect the caustics they produce themselves in a decent way

The fact is they dont resolve their own reflected light, even at high SL it may look noise free but it will lose intensity
Its great lens setup though, save it for when maxwell can render the mirrors, I pray for that day.

EDIT: I had a go at making a lens and it was quite straighforward (used the C4D oiltank primitive with some scaling of the bowls), and renders fast..
It has to be comped because of second bounce caustics as we know but you can get away with it if all the surrounding surfaces have roughness higher than 67.



It looks better when rendered with maxwell rather than V-ray even if needing to be comped, nice trick, Thanks for this

p.s. backlit thin-sss with emitters causes some strange brightened edge shading, thatys a seperate issue..

[eric nixon]

Noticed some noise coming in at higher sl, but its easy to fix, just remove the reflectance of the lens.

[eric nixon]

Another faster way which works without comping!
The challege now becomes focussing, using a target point for the tubes to aim at, its using 3 'light-tubes' R, G , B. And a weightmapped black stencil for each 'tube cover', with the corresponding RGB channel as weightmap.

5mins sl14



To clarify; there are no lenses in this scene.. you can see the focussing issue from the rgb fringing esp noticable at the white border.. the white border is from the gap I left between the light-tube boxes and the weightmapped-cover.