OL2.SEP– RM SIGGRAPH
SIGGRAPH moves toward software
The latest in rendering engines, portable libraries, and displays shared
space at SIGGRAPH '92 with virtual reality, scientific experiments, and
movies made for entertainment
Some of the highlights at SIGGRAPH '92 included improved virtual reality,
multivendor displays using PEX to transfer 3D graphics over a network,
graphics benchmarking, and a clear explanation of fractal image
compression. A hardware engine for texture mapping adds zing to high-end
Iris workstations.
Approximately 30,000 attendees of SIGGRAPH '92 visited about 270 booths
on the trade show floor at Chicago's McCormick Place, and attended courses
and paper sessions in rooms nearby. Only one semiconductor vendor,
Philips/Signetics, introduced a chip (see page 66). A large part of the
show was devoted to visual software, the overwhelming majority of which
was displayed on Silicon Graphics hardware. One example was in the annual
juried animation show, which included the Michael Jackson video, “Black or
White,” displaying one use of “morphing” software. A dancer, shown as a
head shot, turns a profile to the camera and comes back with another face,
often of a different race or gender. A few other topics that might be of
interest: * Virtual reality * Workstations * Graphics benchmarking *
Fractal image compression The annual show does not deal with EDA, so our
readers are not likely to be in the market for much of what is here.
However, they might well learn how workstation capabilities can be stretched
for other uses.
Virtual reality Virtual reality takes a number of forms. Progress in
helmet displays is slow, because there is no mass-market item to drive
development of the tiny flat-panel screens needed to make them.
Three-dimensional computer-generated images themselves can be very
convincing (see photo), but getting inside them is difficult. A system
shown by Virtual Reality Group has 1-in. CRTs in a 3-lb helmet. It was
claimed to be capable of 1,280 x 1,024 resolution, but was showing 640 x
480 monochrome. At $100,000, it is cheaper than anything comparable two
years ago, but still disappointing. More satisfying was an enclosure
called the Cave, with four 10-ft square back-projection panels forming a
floor and three walls.eight people at a time could stand inside.
Polarized glasses showed alternate-image stereo with stunning effect.
Attendees could walk inside molecular models and other images. A
surprisingly three-dimensional appearance was achieved without glasses in
a planetarium dome. A vector-drawing projector drew the Mandelbrot set
with successive iterations coming out of the plane instead of changing
color. The attractors became polygonal-helical tubes, into which the
audience could be zoomed. Silicon Graphics showed a practical use for
virtual reality. An all-around sonic scan from a submarine could construct
an image of the world outside. This would allow an operator inside to
steer between or over obstacles in an environment where light is not
available.
Workstations Although all the major workstation vendors had
large booths at SIGGRAPH '92, only Silicon Graphics had a pervasive
presence in third-party booths throughout the show floor. (remember, this
is a show primarily for industrial design, animation, fluid dynamics,
earth sciences, and other visual professions requiring 3D performance. EDA
is nowhere). Machines from the low-end Iris Indigo up to multiprocessing
Iris 4D were everywhere, powering virtual reality displays and animation
systems. In its own booth Silicon Graphics showed the Reality Engine, a
three (or more) board add-on for the high-end Iris 4D. The accelerator
puts texture-mapping into hardware, and leaves SGI alone in its chosen
market–those who really need it. Hewlett-Packard and Digital Equipment
show impressive speed in 3D rendering now, but without textures their
displays look bland compared to SGI. Sun Sparcstations were in the Du
Pont Pixel booth, showing Du Pont's port of Open GL, the licensable
library from Silicon Graphics, to Sparc. Du Pont announced a license
agreement with SGI making Iris Explorer available on the Sparc. It will
ship in January. Digital, IBM, HP, and Sun Microsystems all had huge
booths networked together. This allowed them all to show PEX (PHIGS
extensions to X) images sent from any of them to the others. Digital
showed two prototype machines, based on the new Alpha microprocessor,
running impressive demos. One was a 130-MHz desktop in the same box as a
DECstation 5000 and the other was a 150-MHz deskside box newly designed.
These are developers' machines. Production for end users is expected at
the end of the year. MIL-spec versions of the alpha boxes will be made by
Raytheon for production versions of the JSTARS battle-surveillance plane,
which now carries three militarized VAX 6000s. This represents a big win
for Raytheon, but will hardly sell a huge number of microprocessors.
Intergraph showed a workstation with a 27-in.screen, 2,000 pixels across.
No one else had anything comparable. The company has ported much of its
software to other platforms, including PCs, but still retains a niche for
its own specialized hardware.
Benchmarking graphics A debate over graphics benchmarking centered on
the Picture-Level Benchnmark (PLB), and whether preprocessing should be
permitted to show the ultimate performance a system could reach. The
interests behind the opposing positions were supporters of the Silicon
Graphics Open GL graphics library, vs.supporters of PEX. The PLB rules
allow both literal and optimized running. The literal numbers are useful
to those unable or unwilling to tune an application to a specific
platform. A port to a new machine would normally entail tuning, so a CAE
vendor might be more interested in the optimized numbers, particularly if
they showed dramatic differences. PEX is a system for sending 3D images
over a network to different machines, so literal benchmarks are
applicable. PLBs do not encompass texture mapping now, a process at which
Silicon Graphics is in a class by itself. A benchmark for texture mapping
will be added to the suite in the future.
Fractal image compression Fractal math has been promoted for several
years as an alternative to the discrete cosine transform (DCT) for image
compression. However, the most active promoter in this country, Michael
Barnsley, has declined to publish his algorithms, regarding them as trade
secrets in his software offering. Yuval Fisher, of the University of
California at San Diego supercomputing center, has published his image
compression work. A lucid explanation of the principles appears as
Appendix A, pages 903 to 919, of a new book,
New Frontiers of Science
umlaut>rgens, and Dietmar Saupe, all of the University of Bremen. Mr.
Fisher contributed the appendix. The publisher is Springer-Verlag.
Fisher's method breaks an image into cells, either rectangular or
triangular, and tests for similarities sufficient to allow grouping. Those
that can't be dealt with, according to a chosen measure of similarity, are
divided into smaller cells. The process iterates until all parts of the
image have been coded. Fisher outlines a method with a prechosen
criterion of accuracy, and one with a prechosen number of iterations. The
first method can produce images that look very good by running a long
time. Like DCT compression, the better the reconstructed image, the worse
the compression ratio. The measure of fit could be any number of things,
like least squares and largest difference over a domain. The 5-lb book
itself is for technically oriented readers who want to know details of how
fractals and chaos theory work. It demands no math beyond what any
engineer would know, but it does have many equations inserted in the text
that must be understood to follow the argument. –Rodney Myrvaagnes
CAPTION:
This stereo image of a carbon 60 molecule (Buckyball) was generated on a
Cray YMP and printed with a Dicomed film recorder. To see the solid image,
cross your eyes so the right eye looks at the left picture and vice versa.
Then converge the central image. (Photo courtesy SIGGRAPH '92.)
Advertisement
