New images and Benchmark updates (long)

Gang,

Here are some details on the geometry and rendering parameters, in case you are interested.

"Open House"
http://mark.technolope.org/transfer/img04_2400.jpg

My research software (http://mark.technolope.org/research/) made the general shape, it tracks vortex sheet motion with an adaptive triangular mesh; the time step used has 39467 tris.
A new tool in my Rocktools package (http://mark.technolope.org/rocktools/) called "rockmarker" (insipred by replmarks) placed 2 open-topped cubes for each triangle (20 new tris over each tri in the above mesh).
I added gensky sunlight and an orange-colored sky distribution via my old (and inaccurate) "skycolor.cal" model.
I put both the mesh and the boxes into the scene.
oconv -n 12 scene03.rad mark_0019_b.rad > scene03.oct (took up 7-800 MB in RAM nonetheless)
rpict -vta -vh 18.75 -vv 18.75 @vp_flat3 -ds 0.1 -dj 0.7 -ab 3 -aa 0 -ad 8 -as 0 -u -ps 1 -x 7200 -y 7200 -t 600 -z img04.zbf -o img04.pic scene03.oct (renders in 200 MB RAM, though)
rpict: 1249498744 rays, 100.00% after 25.675u 0.110s 27.581r hours on bacchus (Athlon XP 2000+, radiance 3.8a Jan 22)
ra_tiff -e -3.5 img04.pic img04.tif
Gimp: Gaussian IIR 3 pixel, reduce 3:1, sharpen 30
The final render will be at 36000x36000, reduced to 7200x7200 for printing at 24"x24", it will take about a month on one Athlon. I have two Athlons working on it right now.

"Boil"
http://mark.technolope.org/transfer/img05_1800.jpg

Same software to make the mesh, even same simulation type (Rayleigh-Taylor instability), this time 1106058 tris.
The mesh was smoothed and normals created with another one of the programs in my "rocktools" package:
rocksmooth vort_geom_0015.obj -s -n -oobj > geom0015.obj
obj2mesh geom0015.obj > geom0015.msh
The scene is simply a closed black cube with the mesh at the halfway point, the view above it, and the whole floor is a glow object.
oconv scene11.rad > scene11.oct
rpict -vtl @vp1 -ds 0.08 -ab 2 -aa 0 -as 0 -ad 8 -lr 20 -lw 0.001 -ps 1 -x 7200 -y 7200 -t 600 -o img05.pic scene11.oct
(the rpict "-u" option wasn't created yet, so I compiled with -DMC)
rpict: 3253638522 rays, 100.00% after 29.674u 0.111s 31.193r hours on zeus (Athlon XP 2500+, ray count flipped 4.2G once)
pfilt -1 -e -3 -x /4 -y /4 -r 0.5 img05.pic | ra_ppm | cjpeg -q 90 > img05_2400.jpg

"Dynamo"
http://mark.technolope.org/transfer/img31_2400.jpg

A different code made the geometry here. I wrote most of it for my job/post-doc for a small company in California. Instead of vortex sheets (my research), this code tracks the motion of vortex blobs/particles. In this case, the flow was confined to stay within a sphere. The snakes in the image are all streamlines, traced along the local velocity at a fixed point in time. There are 170909 spheres and probably just a few more cylinders/cones in total.
The ball is in a square and tall room which is mostly white. A warm-colored area light is near the top, the floor is red, and another light is inside of the ball.
The oconv must have taken up all 950 MB available on my best machine. If I had more RAM, I'd make bigger stuff.
It was rendered in two parts:
rpict -vtl @vp30 -ds 0.1 -dj 0.7 -ab 3 -aa 0 -ad 4 -as 0 -ps 1 -u -t 600 -x 14400 -vl -0.5 -vv 0.625 -y 7200 -o img31_1.pic scene30.oct
rpict: 2495232691 rays, 100.00% after 31.838u 0.133s 32.122r hours on bacchus (flipped once, took 400 MB in RAM)
The other one (-vl 0.5 -o img31_2.pic):
rpict: 2798392870 rays, 100.00% after 22.147u 0.001s 22.196r hours on zeus (flipped once)
pcompos -a 1 img31_?.pic > img31.pic
ra_tiff -e -1.75 img31.pic img31.tif
gimp it down to 2400^2 to make img31_2400.tif
I used Gimp for resizing because the "-aa 0 -ad 4" action creates some very hot pixels, and I can't seem to get pfilt to lose the
light---it always bleeds into surrounding pixels. So, I drop the pic to 8-bit, and smooth and resize in The Gimp.

Finally, the URL for my Radiance benchmark page is
http://mark.technolope.org/pages/rad_bench.html

Thanks!

Mark

ยทยทยท

On Thu, 26 Jan 2006, Charles Ehrlich wrote:

Hi Mark,

Those images are so kewl! Amazing. But, they just whet my appetite for more information. What is the polygon count? Rendering time? Size of .amb file? Origin of software used to generate the model? Any more?

And, what is the link to your benchmark web site again?

Thanks,
-Chas

====================

Date: Thu, 26 Jan 2006 20:35:13 -0500 (EST)
From: Mark Stock <[email protected]>
Subject: [Radiance-general] New images and Benchmark updates
To: [email protected]
Message-ID:
<[email protected]>
Content-Type: TEXT/PLAIN; charset=US-ASCII; format=flowed

Radiance users,

In this, the final day before the SIGGRAPH Art Gallery deadline,
I would like to share with whomever is interested some of the
pieces that I will be submitting. They were all created in
Radiance (duh) and have been scaled down 5:1 or 6:1. I hope this
counts as "giving back" to the Radiance community in some way.

(these are all temporary URLs and they are all > 1.2 MB JPEGs)

"Open House"
http://mark.technolope.org/transfer/img04_2400.jpg

"Boil"
http://mark.technolope.org/transfer/img05_1800.jpg

"Dynamo"
http://mark.technolope.org/transfer/img31_2400.jpg

In other news, I have finally updated my Benchmark page with
results that were given to me up to 4 months ago. I welcome any
new contributions you may have. Parallel results may have to
wait, as there are unresolved timing issues with the Linux
platform.

Thanks!

Mark