Vague Terrain

[Albrecht Durer / Man drawing a Lute / 1525]

The hallways of many university research facilities are lined with printed posters and powerpoint slides from conference presentations, displayed to promote research to colleagues and potential funders. It was in this context that I first learned of the standard test subjects of digital imaging research, as an artist/tourist wandering through a well-funded lab. And it was the three-dimensional models that first caught my eye.

Research into three-dimensional imaging is the more obviously synthetic branch of computer graphics. Early work on two-dimensional imagery experimented briefly with the composition of original images, before jumping ahead to start from the acquired image and work backwards to invent our common imaging algorithms.

Dimensional graphics, on the other hand, advanced much farther along the synthetic arc, wrestling with the simulation of light and color as portrayed from multiple perspectives across highly abstracted and ideal forms. Only later did dimensional scanning produce reliable subjects for testing fidelity and accuracy. Both the synthetic and more referential test subjects bear examination, and of course the distinction gets interestingly blurred at times.

The question to ask of these and other models in our series is - what makes them good models, and what sort of vision are they good for?

Initial experiments in ray-tracing relied on extremely simple objects - the sphere, the torus, the box. The frequent use of checkerboards puts these experiments in league with much older experiments from Renaissance perspective. Compared to those earlier pictures, however, these tests display less interest in the dramatic revelation of depth than in the potential of surface as mirror. The best ray tracing demos turned light's particles into a traveling swarm that smartly reassembled itself on any surface - only to reveal other surfaces. Here we see something of a return to the theories of medieval optics, in which pieces of light and color travel about and lodge themselves onto the surfaces of the world. Where light is reducible to math, it seems to follow that all the world should be a screen. And of course a curved screen would demonstrate the plasticity of the data better than a planar one.

To digress a moment, I can't not mention one of the weirdest (if predictable) demos of this path of "reflection mapping research" - that of the pioneering 1985 CGI video called "Interface," in which a woman kisses a CGI silver robot to demonstrate good reflective algorithms. The above still is borrowed from this site, but follow the link to watch the short video as well.

[Sutherland's bug, in process and product]

The eventual need for for a more verifiable and recognizable virtuality led to the first 3D test subjects based on extant objects. Here we see something more akin to linear perspective's trope of the lute, a complex object chosen for how slight changes in orientation reveal the plasticity of space, rather than the plasticity of light. Ivan Sutherland had his students measure out and translate his old VW bug, but the most famous and persistent test subject of this stage was the "Utah Teapot."

[The original teapot in the Computer History Museum, and a typical 3D image]

The history of this ubiquitous virtual object is told elsewhere. Of interest to me in the story is how the teapot's origins provide one of science's occasional glimpses into the role of women as "domestic support" to male researchers. It was reportedly Sandra Newell's suggestion over tea that led Martin Newell to head off and measure the German Melitta urn into a set of 306 XYZ data points. The labs of modern science, especially those in the history books, are of course overwhelmingly male, and one of the few ways women enter the picture is through their role as quasi-muses, hosting dinners and providing spaces of meditation for inspirational breakthroughs. Sandra may well be a professional in her own right, but accounts of her role in the teapot's story place her in the familiar domestic role. Others are thankfully telling the story of these gendered spaces - such accounts should not miss these not infrequent accounts of spousal influence on the lab.

According to researchers, the teapot's advantages as a test subject are that it is recognizable, contains both convex and concave surfaces, and even "self-shadows," all without growing too complicated as a dataset. Early algorithms developed around this object and the VW demonstrated the interpolation of curves based on angular datasets. Later, researchers tried out their various image-mapping techniques on the thing, or rendered it in dubious materials.

It's interesting to think of this teapot's origins as a form on a potter's wheel, where centrifugal force does the work of smoothing out the work of the hands. Off the wheel, science's first smoothing techniques replace force with repetition. Set a curve, turn and repeat it enough times to assume a smooth line in the eyes of the viewer. How many points to measure to get a good dataset? How many facets to leave to create a convincingly smooth plane? Repetition leads to fluid realism. Very futurist.

[Umberto Boccioni / Development of a Bottle in Space / 1912]

And repetition is exactly what researchers delivered. To look at the endless iterations on this form is to see in the research an anticipation of the manufacturing and design process. SIGGRAPH even issued a "Call for Teapots" for its annual conference in 1989. The comparison to image macros is again tempting, as there is a semblance of an emergent, cybernetic intelligence in the sheer mass of variation. But a mere eBay search for white ceramic teapot shows just how limited this iteration can be. If some recent design software is looking to fix that, we still have to deal with how the system is trained. This I'll cover next time, when I turn to cows, bunnies, buddhas, and the 3d object database.

Some jokingly referred to the teapot as a teapotahedron, allying it with platonic solids, but we never actually see this primal material accumulate into anything. Unlike other basic geometrical forms, the teapot only gets varied in itself, more like a contemporary vinyl toy than a primal substance.

The teapot's lasting legacy on our 3d-imaging technologies is probably in lending an ideal scale to the virtual encounter. If more of these renderings had been created from the perspective of an ant on the table, we'd experience a much less satisfactory resolution of planes into curves. Where linear perspective's early experiments looked to create a magical space that might envelop the whole body, practice with the teapot appealed primarily to the hand. A virtual object would be only as real as it would appear from arm's length.

Footnote: Among the best compilations of information on the subject of this series is the website of Professor Wayne Carlson at OSU. I also recently came across a resonant blog entry by artist Kevin Zucker.

Additional installments of this series: Part One, Part Two.