Subject: | |
From: | |
Reply To: | |
Date: | Fri, 24 Feb 1995 12:55:52 -0800 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
Looking at Daniel Chin's message, I realize that I forgot to
provide any performance info on VolPack. Here is some, for
a 256x256x84 bytes volume of an MRI head: times are in seconds for
one rendering, with different pre-processings of the volume
(none/ simple/ full)
sgi r3000/ irix-4.0.5: 3.2/ .7/ .3
sgi r4000/ irix-5.2: 2.3/ .5/ .3
dec 5000/ ultrix-4.2: 10.3/ 1.8/ .9
sparc1+/ sunos4.1.2: 18.3/ 2.9/ 2.2
sparc20/ sunos4.1.3: 2.5/ .6/ .3
(note 1: it would be interesting to know performances on a fast, cheap NT box
or on a PowerMac; has anyone installed VolPack on these platforms ?)
(note 2: a parallel version of the algorithm can render at video rates
on an sgi MP challenge server)
On my r3000 system, with 256^3 volumes, I experience timings of
6 to 20 seconds per frame for complex scenes. This can be slightly
slower than VoxelView on a VGX, but then the lights & materials settings
are much more elaborate, and so the image quality-information content
is better.
One must realize that such speed comes at the cost of:
1) an optional volume pre-processing time
(pre-processing time being a fixed cost, one uses it only
when a large number of frames will be needed with the same
voxels/edges classification, usually more than 30 or 60)
2) a larger than usual use of RAM, but is not RAM cheaper
than software and graphics pipelines?
Correction: VolPack is by Philippe Lacroute and Marc Levoy.
--Jean-Pierre
|
|
|