Michael F. Deering

Geometry compression

This paper introduces the concept of Geometry Compression, lowing 3D triangle data to be represented with a factor of 6 to times fewer bits than conventional techniques, with only slight los es in object quality. The technique is amenable to rapid decomp sion in both software and hardware implementations; if 3D rend ing hardware contains a geometry decompression unit, applicat geometry can be stored in memory in compressed format. Geo try is first represented as a generalized triangle mesh, a data s ture that allows each instance of a vertex in a linear stream to sp ify an average of two triangles. Then a variable length compress is applied to individual positions, colors, and normals. Delta com pression followed by a modified Huffman compression is used f positions and colors; a novel table-based approach is used for mals. The table allows any useful normal to be represented by 18-bit index, many normals can be represented with index deltas 8 bits or less. Geometry compression is a general space-time tr off, and offers advantages at every level of the memory/interco nect hierarchy: less storage space is needed on disk, less trans sion time is needed on networks.

High resolution virtual reality

I define the lower layers of Virtual Reality to be: the highly-accurate, real-time simulation by computer of the interaction of the physical world with human senses. My focus is on the visual system, the talk will desceribe the techniques used to perform this simulation in several running systems at Sun microsystems. These include: correct perspective viewing equations, correcting for the optics of both human includes details of the Virtual Portal, a 1K x 2K walk-in virtual display device.

The triangle processor and normal vector shader: a VLSI system for high performance graphics

Current affordable architectures for high-speed display of shaded 3D objects operate orders of magnitude too slowly. Recent advances in floating point chip technology have outpaced polygon fill time, making the memory access bottleneck between the drawing processor and the frame buffer the most significant factor to be accelerated. Massively parallel VLSI system have the potential to bypass this bottleneck, but to date only at very high cost. We describe a new more affordable VLSI solution. A pipeline of triangle processors rasterizes the geometry, then a further pipeline of shading processors applies Phong shading with multiple light sources. The triangle processor pipeline performs 100 billion additions per second, and the shading pipeline performs two billion multiplies per second. This allows 3D graphics systems to be built capable of displaying more than one million triangles per second. We show the results of an anti-aliasing technique, and discuss extensions to texture mapping, shadows, and environment maps.

HoloSketch: a virtual reality sketching/animation tool

This article describes HoloSketch, a virtual reality-based 3D geometry creation and manipulation tool. HoloSketch is aimed at providing nonprogrammers with an easy-to-use 3D “What-You-See-Is-What-You-Get” environment. Using head-tracked stereo shutter glasses and a desktop CRT display configuration, virtual objects can be created with a 3D wand manipulator directly in front of the user, at very high accuracy and much more rapidly than with traditional 3D drawing systems. HoloSketch also supports simple animation and audio control for virtual objects. This article describes the functions of the HoloSketch system, as well as our experience so far with more-general issues of head-tracked stereo 3D user interface design.

Leo: a system for cost effective 3D shaded graphics

A physically compact, low cost, high performance 3D graphics accelerator is presented. It supports shaded rendering of triangles and antialiased lines into a double-buffered 24-bit true color frame buffer with a 24-bit Z-buffer. Nearly the only chips used besides standard memory parts are 11 ASICs (of four types). Special geometry data reformatting hardware on one ASIC greatly speeds and simplifies the data input pipeline. Floating-point performance is enhanced by another ASIC: a custom graphics microprocessor, with specialized graphics instructions and features. Screen primitive rasterization is carried out in parallel by five drawing ASICs, employing a new partitioning of the back-end rendering task. For typical rendering cases, the only system performance bottleneck is that intrinsically imposed by VRAM.

FBRAM: a new form of memory optimized for 3D graphics

FBRAM, a new form of dynamic random access memory that greatly accelerates the rendering of Z-buffered primitives, is presented. Two key concepts make this acceleration possible. The first is to convert the read-modify-write Z-buffer compare and RGB&agr; blend into a single write only operation. The second is to support two levels of rectangularly shaped pixel caches internal to the memory chip. The result is a 10 megabit part that, for 3D graphics, performs read-modify-write cycles ten times faster than conventional 60 ns VRAMs. A four-way interleaved 100MHz FBRAM frame buffer can Z-buffer up to 400 million pixels per second. Working FBRAM prototypes have been fabricated.

The HoloSketch VR sketching system

Virtual Reality (VR) systems, on the other hand, are distinguished by using real-time updates of the user’s head orientation and position to re-draw (often in stereo) 3D images in real time. When done properly the effect is very much like that of a hologram; the virtual object seems stabilized in space, users can move their heads around it to one side and see the object in profile. Unfortunately, the ultra-low resolution and extremely distorted optics of most head-mounted displays (HMDs) make most VR systems unsuitable for fine manipulation of 3D objects. Nevertheless, some 3D editing systems have been built for the HMD environment [1]. Because of the resolution limits, the use of these systems is mostly confined to simple positioning of objects already created outside of the VR environment, although actual creation of new objects is possible with some effort. We created HoloSketch based upon the supposition that if a sufficiently high-quality/high-accuracy VR display system were used, a VR-based 3D object editor should be both easier to use and more productive than traditional 2D projection systems.

A photon accurate model of the human eye

A photon accurate model of individual cones in the human eye perceiving images on digital display devices is presented. Playback of streams of pixel video data is modeled as individual photon emission events from within the physical substructure of each display pixel. The thus generated electromagnetic wavefronts are refracted through a four surface model of the human cornea and lens, and diffracted at the pupil. The position, size, shape, and orientation of each of the five million photoreceptor cones in the retina are individually modeled by a new synthetic retina model. Photon absorption events map the collapsing wavefront to photon detection events in a particular cone, resulting in images of the photon counts in the retinal cone array. The custom rendering systems used to generate sequences of these images takes a number of optical and physical properties of the image formation into account, including wavelength dependent absorption in the tissues of the eye, and the motion blur caused by slight movement of the eye during a frame of viewing. The creation of this new model is part of a larger framework for understanding how changes to computer graphics rendering algorithms and changes in image display devices are related to artifacts visible to human viewers.

The SAGE graphics architecture

The Scalable, Advanced Graphics Environment (SAGE) is a new high-end, multi-chip rendering architecture. Each single SAGE board can render in excess of 80 million fully lit, textured, anti-aliased triangles per second. SAGE brings high quality antialiasing filters to video rate hardware for the first time. To achieve this, the concept of a frame buffer is replaced by a fully double-buffered sample buffer of between 1 and 16 non-uniformly placed samples per final output pixel. The video output raster of samples is subject to convolution by a 5x5 programmable reconstruction and bandpass filter that replaces the traditional RAMDAC. The reconstruction filter processes up to 400 samples per output pixel, and supports any radially symmetric filter, including those with negative lobes (full Mitchell-Netravali filter). Each SAGE board comprises four parallel rendering sub-units, and supports up to two video output channels. Multiple SAGE systems can be tiled together to support even higher fill rates, resolutions, and performance.

Explorations of display interfaces for virtual reality

Three alternatives to the traditional head mounted virtual reality display are described. One configuration is the Virtual Holographic Workstation, an external stereo CRT viewed by a user wearing head tracking stereo shutter glasses. Another is the Computer Augmented Reality Camcorder, where virtual objects are composited onto live video using six axis tracking information about the location of the video camera. The third system is the Virtual Portal, where an entire room is turned into a high-resolution inclusive display by replacing three walls with floor-to-ceiling rear projection stereo displays. Details of the systems and experiences and limitations in their use are discussed.<<ETX>>