(That's the ACM standard web page copyright notice, which they request be used for any web republication of papers that appeared in ACM journals or conferences. However, it seems pretty reasonable for general use, even for those papers that were not ACM publications)
Available in Adobe Acrobat format.
Describes a shading language compiler that can automatically produce reduced level-of-detail versions of a complex shader. The method used is analogous to edge collapse operations in geometric level of detail, choosing between the possible replacements of one or more local operations with a simpler one based on an estimate of the error introduced.
Available in Adobe Acrobat format.
At each stage of model simplification, the successive mapping technique creates a piecewise linear mapping between points on the more complex model and points on the simplified model. Using such a mapping, it is possible to automatically generate levels of detail from an original polygonal model with guaranteed error bounds. These error bounds include both displacement normal to the surface and displacement tangential to the surface (swimming). The mapping can also be used to map surface properties (e.g. colors or texture coordinates) onto the simplified surface.
Available in Adobe Acrobat format. The presentation is available in HTML format.
We present an algorithm that enables anisotropic texturing on any current MIP map graphics hardware supporting MIP level biasing. The algorithm computes anisotropic filter footprint parameters per vertex. It constructs the anisotropic filter out of several MIP map texturing passes or multi-texture lookups. Each lookup uses MIP level bias and perturbed texture coordinates to place one probe used to construct the more complex filter profile.
Available in Adobe Acrobat format. The presentation is available in HTML format.
Interactive programmable shading is possible on existing graphics hardware. We treat a single OpenGL rendering pass as a SIMD instruction, and compile the procedural shader into multiple rendering passes. To demonstrate the power of this technique, we have created two shading systems. One compiles a constrained shading language to run interactively on current hardware. The second supports the RenderMan Interface and Shading Language on a software implementation of OpenGL with two extensions. The extensions are color range, providing extended range and precision data types, and pixel texture, using computed framebuffer results as texture coordinates.
Available in Adobe Acrobat format. There is also a web page for this work.
Use a form of environment map to render with an arbitrary isotropic BRDF in an arbitrary environment on graphics hardware. The maps are view-dependent, so to allow viewer motion by capturing or generating maps for several viewpoints. Several of these maps are warped in reflection space and blended to give a new map to use for rendering the scene.
Available in Adobe Acrobat format.
Provides details of the shading system on the PixelFlow graphics machine. PixelFlow supports procedural shading using language similar to the RenderMan shading language. This paper focuses on the issues and optimizations necessary to allow procedural shaders to run and run quickly on an interactive graphics system.
Available in Adobe Acrobat format.
Presents technique for preserving object appearance during geometric simplification. Even with simplification techniques that maintain tight bounds on the deviation between the original and simplified surface, the shaded color can differ drastically. Texture maps and normal maps are created (these can be derived from the original surface representation), then applied to the simplified surface. Since the textures are filtered independently from the geometry, they can retain shading and color fidelity even on heavily simplified models. To use the maps, texture coordinates must be maintained on the simplified surface. To avoid artifacts, a new texture deviation metric is added to the simplification constraints.
Available in html and Adobe Acrobat formats.
Presents a new triangle scan conversion algorithm that works entirely in homogeneous coordinates. By using homogeneous coordinates, the algorithm avoids costly clipping tests which make pipelining or hardware implementations difficult. The algorithm handles clipping by the addition of clip edges, without the need to actually split the clipped triangles or compute vertex positions and shading values at the clipping plane.
Available in Adobe Acrobat format. There is also a web page for this work.
At each stage of model simplification, the successive mapping technique creates a piecewise linear mapping between points on the more complex model and points on the simplified model. Using such a mapping, it is possible to automatically generate levels of detail from an original polygonal model with guaranteed error bounds. These error bounds include both displacement normal to the surface and displacement tangential to the surface (swimming). The mapping can also be used to map surface properties (e.g. colors or texture coordinates) onto the simplified surface.
Available in html and Adobe Acrobad formats. The presentation slides are also available in Adobe Acrobat format.
It will soon be possible at interactive rates to have shading effects that would have been called Photorealistic only a few years ago. We present a discussion of the hardware and software system requirements, some tricks for time and space efficiency, and details of a partially complete "proof by example" on the PixelFlow system.
Bowling animation reminiscent of the
cover of the RenderMan companion. Rendered on a PixelFlow simulator,
conservative timing estimates (detailed in the paper) show that
PixelFlow will be able to rendender this animation at 30 frames/second.
Available in html format.
Latency in the graphics system disrupts the "illusion of presence" we attempt to achieve in head mounted display and virtual environment applications. It is impossible to entirely get rid of the system latency. Head motion prediction helps, but for accurate prediction the latency must still be as small as possible. We examine latency and present a couple of ways we have tried to reduce it. The first is uses alternate rendering software (called SLATS) for the Pixel-Planes 5 graphics hardware. SLATS is optimized for low latency instead of high polygon rendering rate. The second is just-in-time pixels. Just-in-time pixels uses different transformations for every scan line (in the limit it could be every pixel), allowing the possibility of latencies of less than a frame time.
Available in Adobe Acrobat format.
Presentation of 3DM, a fairly early HMD modeler. 3DM is modeled after MacDraw and remains a popular demo in our lab. In hindsight, I think we didn't emphasize enough some of the unique user interface techniques we explored. The web seems as good a place as any to relate a couple thoughts I've had on 3DM's interface.
Not available online.
Describes system which simulates plastic injection molding on a Cray and then presents an interactive visualization on a high-end graphics machine.
I have a separate page with my dissertation in compressed postscript and Adobe Acrobat formats.
Presents a decomposition of the interactive graphics pipeline into procedural stages. Each stage has the potential to be replaced by a procedure written in a high-level language by a user of the graphics machine. Includes details and lessons from the PixelFlow graphics machine, where some of the stages were implemented.
Available through IBM patent server in html format.
Use of interactive 3D computer graphics in plastic injection mold design for visualizing fluid flow simulation results.
Available in Adobe Acrobat format.
Describes level-of-detail shaders that can automatically reduce their complexity to maintain interactive rendering. Describes a system using level-of-detail building blocks for common shading functions like bump mapping or BRDF approximation, with an additional level-of-detail parameter that controls their level, from full accuracy to a single constant color. Also describes a framework that may be used for future systems capable of automatically simplifying complex shaders.
Available in html and Adobe Acrobat formats.
Addresses the question of how to separate the sampling of geometry (at vertices) from the sampling of normals. This is done with a bump-map like normal map. However, as is true for bump maps, this is prone to aliasing. We develop a simple shading model that incorporates both normal direction and shading information. This model is unique because, through simple MIP-map filtering, it can provide a smooth transition from variations in surface normal to changes in shading.
Available in html and Adobe Acrobat formats.
Explores the issues associated with supporting arbitrary procedural primitives in graphics hardware. The primitives are defined with a machine-independent language similar to the RenderMan shading language, and called using extensions to OpenGL.
Available in Adobe Acrobat format. There is also a web page for this work.
Tech-report version of the IEEE Visualization '97 paper.
Available in Adobe Acrobat format.
An experimental rendering program running on Pixel-Planes 5 is described. This system is designed to present the minimum possible image generation latency for a head-mounted display application. While limited in the number of polygons it can handle, it uses a constant single NTSC video field time in display processing for two stereo views. This is a detailed account of the SLATS rendering system described "Combatting Rendering Latency".
Available in Adobe Acrobat format.
The idea is presented of a user immersed in a virtual world with holes through which they see the real world. Before the writing of this paper, augmented reality systems tended to show a view that was mostly real-world with a few small virtual objects, though both paradigms are seen today. The mostly-virtual model has the advantage that simple models of real-world objects (rendered in black or the proper color for chroma-keying) can allow correct occlusion between the handful of objects the computer and the computer graphics elements. Possibilities include being able to see the hand and mouse, computer keyboard and monitor, or a window to the rest of the room.
Details of the hardware implementation detail is also discussed. Most of this part is probably not all that relevant for newer head-mounted displays, but may be of interest to some.
Available in Adobe Acrobat format.
Phong shading (linear interpolation of normals, not to be confused with Phong lighting) fixes the Mach banding problems suffered by Gouraud shading, but still suffers from gross artifacts from errors between the linearly interpolated normals and the normals of the underlying surface. These can be seen in areas of high curvature and near curves of inflection. We present a differential geometry-based explanation of the problem and suggest using a texture map with normals of the underlying surface if possible.
A brief mathematical aside: several reviewers pointed out that our table of spline patch degree vs. surface normal degree had a lower degree for the surface normal than suggested by Shantz and Chang's paper in SIGGRAPH 1988. Since the surface "pseudo-normal patch" must be renormalized anyway, there are multiple valid representations. Our representation did achieve a lower degree than Shantz and Chang, but I can't prove that it is of the minimal possible degree either. I'd be interested if anyone has a proof for the lower bound of this.