 |
|
Cornell's Program of Computer Graphics--long considered the birthplace
of architectural rendering--promises to change the way all future designers
work.
By Jonathan Ringen
June 2002
 |
|
Time & Light |
|
Click here to view a timeline of Cornell's Program of Computer Graphics (PCG) greatest hits over the years. |
|
|
|
 |
|
|
 |
|
Then and now: this 1972 3-D graphic of I.M. Pei's Johnson Art Museum at
Cornell (left) is one of the first ever architectural renderings. The
RenderMan software used to create photorealistic effects shots, like
this one from Pearl Harbor (above), is that rendering's direct
descendant. Thirty-one out of the last 35 films nominated for the Best
Visual Effects Oscar were done with RenderMan.
Photos: Left, courtesy Cornell Program of Computer Graphics; above; © 2001 Touchstone Pictures, courtesy Industrial Light + Magic
|
Computer Science 417: Computer Graphics is cross-listed in Cornell's course
catalog as "Arch 374," but looking around the lecture hall one
doesn't get the impression that many aspiring architects have registered
for the class. The endearingly nerdy group of about 150 is comprised of
exactly the aesthetically disengaged types you would expect to find
studying computer science: most are in sneakers and sweatshirts; no more
than ten or fifteen are women. But what these undergraduate scientists
are being taught--how to make computer models of the way light behaves when
it hits a surface--is the central component of work being done on architectural
modeling and rendering at the university's Program of Computer Graphics
(PCG), work that in the next five years will fundamentally change the
way all designers do their jobs.
Professor Donald Greenberg, who has run the PCG since he founded it nearly
30 years ago, teaches the light-reflection class himself. Surveying
the drowsy students tiered in front of him, he says: "A number of years
ago I taught a class in Beijing, China, and I told a joke. There was simultaneous
translation, and everybody laughed. Afterward, I asked my translator how
he had translated it. He turned red and said, 'I told them American professor
told joke--everybody please laugh.'" Having elicited a round of genuine
laughter with the story, Greenberg responds with typical enthusiasm ("Okay,
now the class is alive!") and, brandishing a laser pointer, launches
into a demonstration of the physical principles behind the behavior of bouncing
light. He darkens the room and shoots his laser at a black-and-white checkerboard-patterned
surface, noting that the white squares reflect and the black squares
absorb. The laser's next victim, a mirror, is so reflective that a
red dot is clearly visible on the back wall of the large room. Then he takes
aim at a piece of glass--which, he points out, both reflects and refracts--and
finally at some brushed aluminum, a material with reflective properties
that vary depending on the orientation of the scores in its surface.
|
|
 |
|
 |
|
 |
A physically accurate rendering of Rhodes Hall's glass-enclosed
stairwell using direct illumination (left), and global illumination
(middle). The globally illuminated rendering takes into account the way
the light bounces from surface to surface, making it more believable as
well as a better predictor of what the built space (right) will look
like. Created at Cornell, global illumination is now included in
off-the-shelf rendering software like 3D Studio VIZ.
Photos: Courtesy Cornell Program of Computer Graphics
|
Greenberg's making two basic points: that to model light you need to understand
its behavior, and that light behaves differently when confronted with each
of the world's bewildering variety of materials. Look around the room you're
sitting in and count the surfaces. Then identify all of the light sources.
In even a simple room with just these two parameters you've already defined
an environment of considerable complexity. If you wanted to make a computer
rendering of that environment--not just interpret it the way an artist would
but really reproduce its physical properties--you'd need to know precisely
how each of the surface materials responds to each of those particular lights.
Natural light from a window has a different wavelength than an incandescent
bulb or a fluorescent tube; the angle of incidence (that at which the
light hits each surface) has to be taken into account. If this sounds a
little esoteric, it's not. This is how a camera--or your eyes, for that
matter--works, gathering waves of light as they bounce off of objects. Getting
the light right is the single most important part of making simulations
that look, and act, like the real world. In renderings, light is everything.
|
|
 |
Richard Meier critiques undergraduate design studio students using
Architectural Studio, soft-ware that allows collaboration over the
Internet. Meier, who is in New York, sees the same images as the
students in Ithaca and can sketch on top of their plans using a stylus
and tablet.
Photos: Courtesy Cornell Program of Computer Graphics
|
Greenberg is undeniably a visionary. Trained as an architect and engineer,
he first wrote 3-D design programs back in the early 1960s, at a time
when computer graphics barely existed. Working with the engineering firm
of Severud Associates, in New York, on the St. Louis Gateway Arch and Madison
Square Garden's roof, Greenberg found the tools that he was using awkward.
"I didn't know graphics was going to be a field," he says.
"I just didn't like looking at numerical output, so I started to write
some graphics programs to display the stuff in three dimensions." In
1965, upon returning to Cornell to teach a course in computer graphics,
Greenberg attended a lecture by Rod Rougelot, a college friend of his who
had gone on to head General Electric's space-flight simulation lab.
"After the lecture we talked, and in a very short amount of time I
had access to GE's laboratory from five in the afternoon until eight
in the morning," he says. "I'd go up to Syracuse with fourteen
architecture students, and we made a [computer animated] movie of the history
of how Cornell was built up." In 1972 Greenberg submitted a proposal
to the National Science Foundation in which he presented a wide variety
of potential applications for computer graphics including models of the
planetary system (with Carl Sagan), the pollution of lakes, the spatial
distribution of deep earthquakes--as well as urban modeling and architecture.
Grant in hand he launched the PCG in 1973. Thirty years later he remains
at the forefront of a field he essentially created. "I don't know
of anyone doing work that's as technologically advanced as his," Greenberg's
Cornell classmate Richard Meier says. "Because of his architectural
training he doesn't just understand technology--he knows what architects
need, and knows how to develop technology that addresses those needs."
|
|
 |
Above, physically accurate automobile renderings give designers a much
better idea of what a full-scale car will look like than less
sophisticated images did.
Photos: Courtesy Cornell Program of Computer Graphics
|
"The major ongoing project we have is photorealistic rendering,"
Greenberg says, drawing diagrams on a large whiteboard in his comfortable
office in Gwathmey Siegel's 1990 Frank H. T. Rhodes Hall, Cornell's
main computer-science building. "There are two goals. We've got to
be physically correct and perceptually indistinguishable. So if you have
an accurate model [a set of data that describes an environment, such as
a room], our rendering will really look like that environment." In
recent years Greenberg has added a third goal: the ability to render in
real time--30 frames per second in computer terms--allowing the viewer to
"walk" through a space, going wherever he or she pleases. These
are all things that, theoretically, Greenberg and his colleagues know how
to do. The obstacle is the finite availability of processing power.
Currently it's possible to do a very complex still rendering or a simplistic
real-time one. But in time that problem will solve itself, because the computer
industry evolves at an exponential rate. Moore's law--an observation made
by Intel cofounder Gordon Moore in 1965 that remains true today--predicts
that the number of transistors that can be crammed onto a chip will double
approximately every 18 months. Pointing to a large rendering of one of Rhodes
Hall's glass-enclosed stairwells, done a year before it was built, Greenberg
says that it took about five hours to compute. That same model now could
be rendered in a couple of seconds. As recently as two years ago, physically
accurate real-time rendering took more than a million times more processing
power than was available. Today Greenberg is down to within just a hundred
times what he needs to render even very complex lighting, materials, and
environments.
|
|
 |
Pixar's RenderMan provides the artists working on movies like
Monsters Inc. amazing control. The wireframe model contains
information about shape, movement, and expression; smooth polygons are
added to give the model color and form; and the final rendering adds
lighting, texture, and motion blur.
Photos: Courtesy Disney/Pixar
|
When Greenberg talks about available processing power, though, he's not
referring to the computer on your desk. In a quietly humming room that smells
slightly of ozone is the PCG's cluster computer (funded by Intel): 64 black
computer towers lined up on shelves, each housing two Pentium 4 processors,
and all connected to work as one. But given the rate of innovation in the
industry, it's no more powerful than a typical new desktop computer will
be in five or ten years, making technology that now exists only in
the rarified world of academic and industrial labs accessible to everyone.
What you get when you visit Cornell is a glimpse into the near future, when
much of the lab's current work will be available in a commercial package
that will run on an ordinary PC. There is already precedent for this: the
stairwell, for example, which took five hours to compute on a mainframe
in 1989, can currently be rendered on a consumer computer using Autodesk's
Lightscape--commercially available software developed, of course, by Cornell
grads--in less than five minutes.
What a physically accurate, photorealistic, real-time rendering offers that
a pretty drawing doesn't is predictive power. And predictive power is what
makes Cornell's technology so revolutionary--not just to architects but
to the automobile industry, product designers, Hollywood, plastic surgeons,
even dentists. "Car companies are shifting to digital design to reduce
the number of physical models they deal with, but they still run into situations
where they do the best visualizations they know how to do, mill a full-size
model, and are surprised at the result," says research staff member
Steve Westin, who's working on a project with the auto industry. "If
you do a scientifically better job of making the pictures, how much
does that improve your ability to predict what's really going on?"
|
|
|
BACK TO TOP
