The final render is up on the TC-RTC
http://tc-rtc.co.uk/display/index.html
Under the title, "Systema Metropolis: In the Wild."

The code, including the tree macro, is posted to p.b.s-f.

Finally: Surprise!

How many of you saw *that* final scene coming? :-D

-- 
William Tracy
afi### [at] gmailcom -- wtr### [at] calpolyedu

One time I put my car into reverse while traveling 70 mph on the 
freeway. It was a very exciting learning experience.
     -- seen on Slashdot

Post a reply to this message

I'm playing with this scene and huge resolutions :)

Of course, to render at super-high resolutions, tiled rendering is 
needed (to distribute between computers/cores). But I can't split the 
scene into tiles using the "normal method" (setting +w and +h to the 
full res, and using +sc +ec +sr +er). I tried it with very high 
resolutions (32768 x something) and POV-Ray crashed, or gave out of 
memory error, apparently because it attempts to allocate memory for the 
whole image. However, I successfully managed huge tiled renders with 
zoomin.inc, where I can set +w and +h to the size of a single tile.

#include "zoomin.inc"

#macro Str(N) str(N,0,-1) #end
#declare Rows=4;
#declare Cols=4;
#macro ZoominParams(X,Y)
   concat(
     " +sr",Str(Y/Rows),
     " +er",Str((Y+1)/Rows),
     " +sc",Str(X/Cols),
     " +ec",Str((X+1)/Cols)
   )
#end
#declare Tile = frame_number;
#local X = mod(Tile, Cols);
#local Y = div(Tile, Cols);

Set_Zoom_Area(camera_location, camera_look_at, ZoominParams(X,Y))
Update_Camera_Zoom()

I rendered a 2048x1536 image, using 16 tiles, just for the sake of 
testing. I can't really tell how much time it took, since I kept 
changing some things while the tiles were being rendered, and then 
re-rendered the ones that were left with old parameters.

Area lights and radiosity are disabled. I tested radiosity, and 
confirmed it gets noticeable artifacts between one tile and the other. 
All trees were loaded from linden-lowres5.inc, including bushes, to 
decrease parse time to around 15 seconds.

I just noticed the PNG image is 5.3MB, even after optipng'ing it! JPEG 
got it to 800KB while keeping more than acceptable quality. EDIT: and I 
had to lower quality a bit more; I forgot how much overhead base64 
causes, and the server gave a size limit error. Down to 700KB, retrying.

Post a reply to this message

Attachments:
Download 'inthewild_big.jpg' (680 KB)

Preview of image 'inthewild_big.jpg'

Nicolas Alvarez nous apporta ses lumieres en ce 2008/02/13 12:11:
> I'm playing with this scene and huge resolutions :)
> 
> Of course, to render at super-high resolutions, tiled rendering is 
> needed (to distribute between computers/cores). But I can't split the 
> scene into tiles using the "normal method" (setting +w and +h to the 
> full res, and using +sc +ec +sr +er). I tried it with very high 
> resolutions (32768 x something) and POV-Ray crashed, or gave out of 
> memory error, apparently because it attempts to allocate memory for the 
> whole image. However, I successfully managed huge tiled renders with 
> zoomin.inc, where I can set +w and +h to the size of a single tile.
> 
When rendering at very large dimentions, add "+d" to suppress the prevew window. 
This will save memory.

-- 
Alain
-------------------------------------------------
I believe that banking institutions are more dangerous to our liberties than
standing armies. Already they have raised up a monied aristocracy that has
set the government at defiance. The issuing power should be taken from the
banks and restored to the people to whom it properly belongs.
Thomas Jefferson

Post a reply to this message

Alain escribió:
> When rendering at very large dimentions, add "+d" to suppress the prevew 
> window. This will save memory.
> 

With display enabled, I got an out of memory error. When using -d, I got 
a crash.

zoomin will do :)

Post a reply to this message

Nicolas Alvarez wrote:
> I'm playing with this scene and huge resolutions :)

On the one hand, that is really cool.

On the other hand, all the little mistakes I'd hoped nobody would notice 
are now magnified (several of the machines are levitating slightly ...). 
Eek!

-- 
William Tracy
afi### [at] gmailcom -- wtr### [at] calpolyedu

In Fig. 3.18 we define two local variables, four and twenty (no jokes 
about blackbirds, please).
     -- Jeffrey D. Ullman, _Elements of ML Programming_

Post a reply to this message

> On the other hand, all the little mistakes I'd hoped nobody would notice 
> are now magnified (several of the machines are levitating slightly ...). 
> Eek!
> 

Hey, most people would have faked the trees at the resolution you 
originally used... (fake meaning they wouldn't be meshes with individual 
leaves!)

Anyway, take it as an opportunity to do those little fixes :) It could 
be even more res next time:

http://stuff.povaddict.com.ar.nyud.net/inthewild_tile52-100.png

tile <5,2> (on a 10x10 grid), all trees linden-lowres5, area lights on, 
radiosity off

Post a reply to this message

Nicolas Alvarez wrote:
> Hey, most people would have faked the trees at the resolution you 
> originally used... (fake meaning they wouldn't be meshes with individual 
> leaves!)

I was running out of modeling time, and had the choice of trying to fake 
the trees with a texture on the ground, or using (mostly) pre-modeled 
trees from Povtree. normal{bumps} wasn't cutting it at this resolution, 
so I used Povtree trees.

> Anyway, take it as an opportunity to do those little fixes :) It could 
> be even more res next time:

What are the specs on the machine you're running this on? You've 
inspired me to try a 2048x1536 RSOCP render; I'm finding that I have 
more than enough memory to do it in one shot, but it is taxing my processor.

-- 
William Tracy
afi### [at] gmailcom -- wtr### [at] calpolyedu

And the Format source code tells me that *really* the indicator of an 
error is whether the index has changed value or not. If it has *not* 
changed value, then some of the Format class' innards assume that since 
parsing didn't happen, an error must have occurred. (Pause for 
Buddhist-like contemplation of this leap of logic and faith.)
     -- Laird Nelson, blogging on the "dank recesses" of java.text

Post a reply to this message

> What are the specs on the machine you're running this on? You've 
> inspired me to try a 2048x1536 RSOCP render; I'm finding that I have 
> more than enough memory to do it in one shot, but it is taxing my 
> processor.
> 

AMD Athlon 64 X2 Dual Core 4200+, 2.20GHz
2GB RAM
Running Windows XP SP2 32-bit, so 32-bit POV-Ray as well

I did the 2048x1536 render with 16 tiles, manually making sure both 
cores were always busy with one tile or the other :)

Post a reply to this message

> On the other hand, all the little mistakes I'd hoped nobody would notice 
> are now magnified (several of the machines are levitating slightly ...). 
> Eek!
> 

I'm now seeing trees that are mostly below the landscape surface.

Post a reply to this message

> http://stuff.povaddict.com.ar.nyud.net/inthewild_tile52-100.png
> 
> tile <5,2> (on a 10x10 grid), all trees linden-lowres5, area lights on, 
> radiosity off

http://stuff.povaddict.com.ar.nyud.net/inthewild_tile82-100.png

The rust texture looks pretty good even at that res...

Post a reply to this message