Problem with layered reflection / refraction

Announcements, requests and support regarding the Blender Indigo export script
User avatar
Heavily Tessellated
Posts: 108
Joined: Thu Aug 10, 2006 4:20 pm
Location: Huh?

Problem with layered reflection / refraction

Post by Heavily Tessellated » Sun Sep 02, 2007 12:43 pm

Hi. Can anyone help me figure this out? This is probably not Blender specific, I just gravitate to the Blender topic... move it as you see fit.

The model in question is a Galilean thermometer. It's a basic "studio shot" - 5 meshlights; one overhead, two lateral, two 45* frontal off-axis. The scale is roughly the 1BU = 1m standard; eg. the base is 0.1 BU wide. This is a metro pass, bidir comes out much worse, the bulbs are nearly invisible.

I am getting no reflections / refractions off the inner bulbs nor the fluid inside said bulbs, so figuring it was a CSG issue, I took the column of alcohol-water and copies of the stuff inside and merged it. As you know, Blender boolean sucks, and even MegaBool did not do well here... it took hours of meticiously hand re-facing to make a hollowed out version of the column. The end result was the same, however. You figure any photon passing through just one bulb has thirteen faced surfaces to pass... glass outer -> glass inner -> alcohol-water outer > alcohol-water inner -> bulb outer -> bulb inner -> fluid inner & outer -> bulb / alcohol-water / glass tube inner & outer again. Each sphere should be reflective, as should the fluid. As you can see, they're not. You only see the lights reflected in the outer glass and the backface of said glass - the rest, nothing. Not even the top of the alcohol-water which should be the easiest inner refection, but as you can see it's nearly invisible. (just above the top red bulb.)

I've tried tweaking the color densities from like 0.01 to 10, but all this does is make the transparencies less transparent. It does nothing for making the reflections show up.

Compare this to any photograph of a Galilean thermometer you can find on the web. It just comes out wrong.
Attachments
Galilean-Thermometer-test4.jpg
400x1200
Galilean-Thermometer-test4.jpg (83.3 KiB) Viewed 4030 times
Last edited by Heavily Tessellated on Tue Sep 04, 2007 1:08 pm, edited 1 time in total.

User avatar
zsouthboy
Posts: 1395
Joined: Fri Oct 13, 2006 5:12 am

Post by zsouthboy » Sun Sep 02, 2007 1:03 pm

Looks like a precedence issue to me.

neepneep
Indigo 100
Posts: 413
Joined: Wed Jun 28, 2006 10:21 am

Post by neepneep » Sun Sep 02, 2007 1:05 pm

Did you set the precedence values for each medium?

User avatar
Heavily Tessellated
Posts: 108
Joined: Thu Aug 10, 2006 4:20 pm
Location: Huh?

Post by Heavily Tessellated » Sun Sep 02, 2007 3:19 pm

OK talk to me like I have no idea what this pres-i-dents thing is...

They're all set to 10, from Blendigo's default. I take it that's incorrect. What should I be doing?

If this is something already covered ad naseum just link me to the proper thread.

Thanks!

oogsnoepje
Posts: 35
Joined: Sat Aug 18, 2007 10:20 am

Post by oogsnoepje » Sun Sep 02, 2007 9:52 pm

Finished searching for:
precedence

Total instances found:
3
precedence

Precedence is used to determine which medium is considered to occupy a volume when two or more media occupy the volume. The medium with the highest precedence value is considered to occupy the medium, 'displacing' the other media.

The predefined and default scene medium, 'air', has precedence 1.

type: integer
unit:
restrictions: Should be > 1
Source: [url=http://Documents_and_Settings/Heavily_Tessellated/Desktop/indigo_v09_test6/it_is_there_i_am_telling_ya/Indigo_Manual.pdf]Indigo Manual[/url] <-- you can find more information there

zuegs
Posts: 380
Joined: Tue Jun 27, 2006 5:46 pm
Location: switzerland

Post by zuegs » Sun Sep 02, 2007 9:54 pm

the more a medium is encapsulated by other mediums the higher you need to set the precedence. Let's say outer glass (10), water (11), balls glas (12), liquid inside balls (13), ....
means that medium with highest precedence has proiority at specific location :wink:

User avatar
Kram1032
Posts: 6649
Joined: Tue Jan 23, 2007 3:55 am
Location: Austria near Vienna

Post by Kram1032 » Sun Sep 02, 2007 10:03 pm

If you've 13 different intersections, I guess, you need 14 different precedence settings!

No booleans needed at all :)

you simply model the glass tubes with thickness, then, the fluid, then the glass bubbles, then the air inside (just a smaller version ;)) then the fluid on them and you set the precedence getting bigger and bigger to the inside....

or the fluids get a lower precedence than the glass. Just the air inside the glass bubbles should have higher precedence than the glass bubbles themselves and the liquid even higher precedence :)

Btw: that's the PERFECT precedence test :D

User avatar
Heavily Tessellated
Posts: 108
Joined: Thu Aug 10, 2006 4:20 pm
Location: Huh?

Post by Heavily Tessellated » Mon Sep 03, 2007 6:33 am

Kram1032 wrote:If you've 13 different intersections, I guess, you need 14 different precedence settings!

No booleans needed at all :)
Ah this is good news, I was considering animating a bulb or three to do a changing temperature scene, but the thought of having to re-boolean each position was going to cause an aneurysm...

With so much transparency it takes a while to get even 1k/pix convergence on four cores, I'll try the increment-by-one method and post the results.

Thanks guys!

User avatar
Heavily Tessellated
Posts: 108
Joined: Thu Aug 10, 2006 4:20 pm
Location: Huh?

Post by Heavily Tessellated » Tue Sep 04, 2007 7:40 am

Hmm. OK. This time I used the non-Boolean'd cut-out version, and set the precedences incrementally inward; eg. glass cylinder 10, alcohol-water column 11, glass bulbs 12, colored fluids 13... the result is better than last time but definitely not correct, I do get the proper inner reflections like the "top" of the alcohol-water column and all 8 reflections of the 4 side lights (backfaces too!! woo!), but they're quite weak, even with the color gain down as low as 0.2... you can barely see the lights on the inner bulbs, with the water and glass at 0.0 color.

Man, but it's very computationally intensive; so this is only resolved to 2000 samples per pixel and took 12 hours.... still, it's enough that if the fluid colors were going to come through, it would have.

Anyway, any ideas where to go from here? Being that it's ~8h each test, I don't want to just keep increasing the color gain randomly till I get something close three months from now.
Thanks.
Attachments
Galilean-Thermometer-test5.jpg
400x1200
Galilean-Thermometer-test5.jpg (99.95 KiB) Viewed 4028 times
Last edited by Heavily Tessellated on Tue Sep 04, 2007 1:09 pm, edited 1 time in total.

User avatar
Kram1032
Posts: 6649
Joined: Tue Jan 23, 2007 3:55 am
Location: Austria near Vienna

Post by Kram1032 » Tue Sep 04, 2007 8:00 am

I'd say, now, your problem is, that you use background light or very uniformal light... to get stronger reflections, you need sharper light that comes from less directions (or has a peak at one of them) - try a studio setup, or something like that :)

User avatar
Heavily Tessellated
Posts: 108
Joined: Thu Aug 10, 2006 4:20 pm
Location: Huh?

Post by Heavily Tessellated » Tue Sep 04, 2007 10:16 am

:P It is a basic studio setup, as detailed in the initial post.

Besides, this should not matter, the strength of the reflection should be the same relative to the light source. If it was just a low-strength blackbody emitter emulating a single candle, the effect should be the same.

OK let's tackle this one bit at a time. Before working on the colored fluids... so why, with the outer glass cylinder and the alcohol-water column set as <rgb>0 0 0</rgb> are they so weak? "They" being the primary surface reflections of the light sources on the bulbs - They should be 99% as strong as the outer glass ones, which they are clearly not. Any idea how to bump them up, when essentially the material is invisible already, it's only the IoR we're seeing?

BTW: In case anyone's curious about what I'm attempting or what they should look like, head over to Flickr: http://www.flickr.com/search/?q=Galileo+thermometer

Sorry if I sound frustrated. But well, I am. Thanks for any suggestions.

User avatar
CoolColJ
Posts: 1738
Joined: Mon Jun 25, 2007 1:47 pm

Post by CoolColJ » Tue Sep 04, 2007 11:44 am

man you need to jpeg your large PNG files ;)

are you using bidrectional MLT? If you have lots of layered glass material then it's the only rendering mode that will render it properly

User avatar
Heavily Tessellated
Posts: 108
Joined: Thu Aug 10, 2006 4:20 pm
Location: Huh?

Post by Heavily Tessellated » Tue Sep 04, 2007 1:20 pm

>>man you need to jpeg your large PNG files

Yea, sorry about that. Not everyone is bandwidth blessed and there's no reason. Fixed. Thanks.

And yep, bidir metro max depth 1000... typical Blendigo default.

Speaking of, could the Blendigo defaults re: Reinhard and exposure_duration be part of the problem? If it didn't take 4 hours to resolve enough to see the bulb edges I'd try every freakin' setting until I got it thru trial-and-error, but it's just too intensive... I'll probably have to make a more simple test scene and get it working, and extrapolate that to the full scene. (one little detail on a window sill of a whole room is bogging me down for days - grr.)

I'll keep playing with it... it is maddening when Indigo output looks less correct than Blendternal. :P

User avatar
CoolColJ
Posts: 1738
Joined: Mon Jun 25, 2007 1:47 pm

Post by CoolColJ » Tue Sep 04, 2007 1:29 pm

bulb edges.....try higher max depth..like the Indigo default of 10000. With multiple glass objects ray bounce becomes really important. Other than that I find glass can keep showing new things for a long time!

Reinhard won't make too much of a difference as it's a post render tone mapping. Exposure duration just determines how the depth of field of the camera will look.

make the image smaller for your tests, remeber if you half your resolution each side, your speed the render up x4 :)
yeah zoom into a trouble spot and use a small image size. 100x100 imgae of glass renders in 15min to 3000 samples on my P4 3.2ghz in MLT mode
You seem to be rendering a lot of empty space around the object, so you can change the aspect ratio and remove the wasted space

you want it to look like this?


Image

User avatar
Heavily Tessellated
Posts: 108
Joined: Thu Aug 10, 2006 4:20 pm
Location: Huh?

Post by Heavily Tessellated » Tue Sep 04, 2007 3:45 pm

aye. spot on.

Post Reply
20 posts

Who is online

Users browsing this forum: No registered users and 19 guests