February 1st & 2nd
[Graphics] [Physics] [Speaker Bios]
This information is subject to change.
|
|
GRAPHICS SEMINAR SCHEDULE February 1st & 2nd [Graphics] [Physics] [Speaker Bios] This information is subject to change. |
The days of drawing predefined static meshes are over. |
The mathematics behind advanced pixel operations, and how to apply those operations to consumer hardware, is fairly well understood these days, both by graphics researchers and by the top game developers. Geometry processing in games, on the other hand, lags far behind in sophistication, even though 3D hardware is advancing in both geometry throughput and geometry features at incredible rates. What is the real mathematical structure underlying the various types of geometry manipulation, including level-of-detail operations, skinning and morphing, set operations, signal processing on geometry, and curved surface representations? How can game developers apply an intuitive understanding of this underlying structure to solve today's and tomorrow's graphics problems in shipping games? Subdivision surfaces, displacement maps, wavelets, voxels, and other technologies all threaten to unify geometry and pixels in the near future under a single mathematical framework. What will this mean for high end game engines? What features of 3D hardware are available or lacking for advanced geometry manipulation? This intensive seminar examines the state of the art in geometry manipulation and processing, exploring both the mathematical side of the problem and the implementation side, on current and near-future 3D hardware. Extended question and answer periods and peer discussion sessions ensure attendees can hone in on specific issues related to their projects. |
| Prerequisites: Graphics seminar attendees are expected to have a complete mastery of game geometry operations, including matrix manipulations, vertex transformation and the like. In addition, familiarity with linear algebra is important, as is complex arithmetic, trigonometry, and calculus. Attendees who understand signal processing before the seminars will derive extra value from the talks. |
Current and Near-Future Consumer Hardware |
| Current and Near-Future Consumer Hardware The Game Technology Seminars try to find the perfect balance between covering mathematical structure and discussing applications for games today. This topic section covers the state of the art as it stands now. Did the hardware guys spend their time in the right or the wrong place? What parts of current and near-future geometry hardware can you make use of when you're writing advanced geometry algorithms for real-time games? This section will set the stage for advancing the state of the game geometry art in the other topics. |
|
The Current State of the Art in Hardware Geometry Performance
and Features, With Peer Discussion
February 1, 2001 9:15 AM to 10:15 AM (1 hour) Brian Sharp, Everybody This survey session will cover the current state of consumer 3D graphics hardware from a features and performance standpoint. A set of performance tests will be disucssed, along with data generated from these tests on an array of current hardware. The talk will segue into a discussion of the state of the art in 3D hardware geometry processing. |
| Survey of Mathematical Features Required by Games, Peer Discussion February 1, 2001 6:00 PM to 7:00 PM (1 hour) Casey Muratori, Everybody As we wrap up this first day of continuous geometry manipulation, we'll come full circle back to implementation and today's game geometry problems. What mathematical and geometric requirements do modern rendering algorithms place on game engines? How can real mathematics help us derive robust and stable answers to questions like: What is the best way to form a consistent tangent space parameterization for dot-product bump map lighting on an arbitrary 3D model? How many bones is enough, and what blending math should one implement? How will programmable vertex hardware affect the flexibility of the geometric pipeline? |
| Geometry as a Mathematically Manipulatable Object While incredibly useful and mathematically sound techniques exist for manipulating 1D signals and 2D images, the same techniques are only now being developed for 3D geometry. Previously ad hoc hacks that only approximated real filtering operations are being replaced by formal methods that enable us to bring the entire field of signal processing to bear on geometry. These techniques will enable a new level of graphics quality as we understand them better and learn to apply them to game graphics. |
| Geometry as a Mathematically Manipulatable Object February 1, 2001 10:30 AM to 10:45 AM (15 minutes) Chris Hecker This lecture gives the rationale for wanting to manipulate geometry in a formal and mathematical way. The benefits and many faces of signal processing are discussed, and it's made clear that if we can only bring the signal processing toolbox to bear on geometry, we'll enable a huge array of interesting operations. |
| Meshes > Subdivision > Wavelets > Signal Processing February 1, 2001 10:45 AM to 11:45 AM (1 hour) Peter Schröder Doing signal processing on meshes is quite different from DSP on regularly sampled 1D signals or 2D images. Meshes are irregularly sampled, topologically complex, sometimes ill-formed, and not what we currently think of as a sampled signal. Subdivision theory gives us a way to formalize meshes as mathematical objects. Wavelets provide a local basis for analyzing the mesh and its frequency and spatial components. Signal processing gives us a huge and well formalized toolbox for performing thousands of different operations on the mesh. This lecture will discuss how this entire process works, and why, and what the issues are. |
| Subtleties of Subdivision February 1, 2001 1:15 PM to 2:15 PM (1 hour) Tony DeRose, Everybody On the surface, subdivision is one of the easiest graphics operations to implement. You simply divide up the triangles and get a new mesh. Delving deeper into what's really going on when we subdivide geometry reveals subtleties that can either help or hinder our understanding and use of the technique. This discussion with Tony DeRose will examine some of these subtleties in an interactive interview and Q&A session. |
| Mesh Signal Processing Algorithms and Applications February 1, 2001 2:15 PM to 3:15 PM (1 hour) Igor Guskov Signal processing on meshes is still in its infancy, especially as far as the game industry is concerned. We're leading the curve by covering this topic now. However, there are applications for mesh signal processing today that we can survey, and there are extrapolations we can make from the vast array of non-geometry signal processing operations that will soon make the transition to geometry. This talk will delve deeper into how the algorithms for digital geometry processing, and then turn to describing and discussing applications of this new branch of graphics. |
| Multi-scale Animation and Geometry Processing February 1, 2001 3:30 PM to 4:30 PM (1 hour) Ken Perlin Multi-scale surface representations enable procedural texture synthesis to be used together with acquired shape data, for powerful and general shape modeling, editing and interaction. The talk will show how these combined techniques can be used at interactive rates, locally and globally over surfaces, to define textures and surface deformations as well as to seamlessly fuse and mix together models with wildly different shape and textural characteristics. An interesting thing about working this way is that it lets you work within different levels of a multi-scale representation, so that designers can work across many levels of scale. The talk will also discuss operators and instructions for future hardware that might be useful for digital geometry processing. |
|
Normal Meshes and Displaced Subdivision Surfaces: Overview,
Comparison, and Discussion as Next Modeling/Rendering
Primimitve
February 1, 2001 4:30 PM to 5:45 PM (1.25 hours) Igor Guskov, Henry Moreton, Everybody This year's Siggraph had two papers covering the same topic: how to represent geometry in an efficient way, and in a way that's amenable to signal processing, multiresolution modeling, and other mathematical operations. Both take advantage of the concept that a 3D model is actually a 2D manifold, and using full 3D information for the model is a waste. How do these techniques compare, both theoretically and in implementation? How do these methods deal with the crappy models that some artists and modeling software produces? Are these techniques really the next rendering primitive? |
| Geometry for Graphics' Sake Doing signal processing on geometry is definitely going to be a big part of graphics technology going into the future, but sometimes you just want a mesh to look pretty on the screen and match your artist's intent. How are we going to represent geometry in games in the near term? Will curved surfaces and continuous level of detail live up to the hype, or will polygons always rule? |
| State of the Art in Curved Surface Representations February 2, 2001 9:15 AM to 10:15 AM (1 hour) Jeff Lander Curved surfaces have a mixed history in the real-time game industry. They're always on the verge of becoming useful and never quite being worth the trouble. What is the current state of curved surface academic research and industry practice, both in other fields and in the game industry? This talk will cover not only broad categories of surface types, but also incremental but important improvements on exising representations. |
| Are Curved Surfaces Ever Worth It? A Visual Complexity Argument February 2, 2001 10:30 AM to 11:30 AM (1 hour) Sean Barrett The conventional wisdom says curved surfaces will be a popular technology in graphically intensive games any year now, just as soon as processors and graphics cards get fast enough. Barrett argues that curved surfaces will never be worth the tradeoffs. Instead of coming at it from the boring old performance viewpoint, he takes a completely novel position and makes the case that you don't actually want curves in the first place. This lecture will raise a number of controversial points about game graphics in general and curved surfaces in particular, and it will smoothly blend into the following peer discussion. |
| Curved Surfaces Discussion February 2, 2001 11:30 AM to 11:45 AM (15 minutes) Everybody This peer discussion will examine the reality of using curved surfaces in shipping products. Are curved surfaces worth it? How are top developers using them, planning to use them, or avoiding them? Compare your experiences implementing curved surfaces in games with other top developers' experiences. Has anyone stayed a fan of curves for more than one product cycle? If you had to implement curves again, how would you do it differently? How did you do it differently the second time? What numerical problems did you run into, and how did you solve them? |
| When is Geometry Important? The Art & Science of LOD Metrics February 2, 2001 12:45 PM to 1:45 PM (1 hour) Jonathan Blow It is becoming commonplace for game geometry systems to dynamically vary the number of primitives in the scene to optimize certain runtime characteristics, like frame rate, or visual quality. These systems make decisions about when to increase or decrease the detail level based on a number of different metrics. How should the programmer choose and tune these metrics? Is there a theoretical structure available to guide us in the choice of a metric? How does the performance overhead of the metric affect the entire system? Blow discusses specific metrics, and derives his own techniques. Theoretical underpinnings and practical implementation details are given equal weight. |
| State of the Art in Alternative Geometry
Representations & Peer Discussion on the Future of Geometry February 2, 2001 5:45 PM to 7:00 PM (1.25 hours) Sean Barrett, Everybody This lecture starts as a report from the fringe. While most of the information today has been practical and useful in the near term, this talk will cover non-traditional geometry representations that are unlikely to make it into your next ship cycle. However, learning about the advantages and disadvantages of alternative representations helps you think outside the box, and possibly helps you be prepared for a potential revolution in the future. This will be an interactive lecture, and we'll discuss the utility of the surveyed representations as they're presented. We'll smoothly blend into a peer discussion of the future of geometry in games. |
| Discrete Geometry Processing Not every operation you want to perform on geometry is a continuous signal processing one. In fact, geometry itself has both discrete and continuous properties. An object has topology, geometry, a graph structure of its mesh data, and other features of interest that are not necessarily best viewed as a continuous signal. Booleans and other computational geometry operations are an important set of manipulations, and will only become more important as the uses of dynamic geometry increases. |
|
Computational Geometry
February 2, 2001 1:45 PM to 2:45 PM (1 hour) Subhash Suri Computational geometers have worked on problems germane to the game industry, and to advanced geometry processing, for decades. Yet computational geometry is one of the least-well-known fields to most high end game developers. This lecture introduces the field using two real world examples--convex hulls and triangulations--to show how computational geometers think about and solve problems, and how they analyze solutions. |
| Randomized Geometric Algorithms February 2, 2001 3:00 PM to 4:00 PM (1 hour) Leonidas Guibas The major intellectual contribution of the area of algorithms to Computer Science as a whole over the past fifteen years has been the introduction of the use of randomization as a fundamental tool of algorithm design. Randomization not only yields some of the asymptotically most efficient algorithms -- it does so through algorithms that are amazingly simple and therefore eminently implementable and practical. In this talk I will try, through a couple of examples, to convey the gist of why randomization has proved so useful in designing geometric algorithms for a variety of problems. Surprisingly , the basic reason is related to the same mathematical structures underlying other areas, such as PAC learning. |
| Numerical Accuracy in Discrete Geometry Operations February 2, 2001 4:00 PM to 5:00 PM (1 hour) Christoph Hoffmann Writing geometric algorithms that are robust in the face of finite numerical precision and sampling errors is incredibly difficult. Anyone who has written geometric intersection or BSP code can attest to this fact. This lecture will cover the state of the art in numerically robust geometric operations, and then discuss how to solve specific common problems with numerical accuracy. |
| Numerical Accuracy Peer Discussion February 2, 2001 5:00 PM to 5:30 PM (30 minutes) Everybody Bring your numerical accuracy problems and have them crushed in the face of intelligent peer discussion between the attendees and speakers. Find out if you're alone in having your point-plane tests give different signs depending on the time of day! |
| Contact info@techsem.com or call us at 310/ 375-6602. |