CS 651: Modern Research in Computer Graphics

List of Papers

Here is a preliminary list of papers you may choose to present. We will add papers to this list as the semester goes on and the interests of the class become clear, and check off papers that we’ve already covered. If a particular paper not on the list interests you (perhaps from one of the videos we've seen), come talk to me and I will consider assigning it or covering it myself.

There is another list of papers (with some overlap) focusing on animation from Dave Brogan’s animation class at http://www.cs.virginia.edu/~dbrogan/CS551.851.animation.sp.2000/justPapers.html.

Mail us a list of your first, second, and third choices for papers you would like to present.

• Evolving Virtual Creatures by Karl Sims (S94, p 15). Not exactly mainstream computer graphics, more the field of "artificial life". Extremely cool anyway. A genetic algorithm is used to evolve virtual critters, both their physical shapes and the neural networks controlling their behavior. Physically-based simulation is used to evaluate the creatures' fitness to do some task: run, jump, swim, etc. Could make a great project.
• Surface Simplification Using Quadric Error Metrics by Michael Garland and Paul Heckbert (S 97, 209). Polygonal simplification is the field in which I did my doctorate and this in my opinion is one of the best simplification algorithms. Simple, fast, and easy to implement, and typically producing high-fidelity results. Implementing this algorithm should make a relatively easy project.
• Rendering Complex Scenes with Memory-Coherent Ray Tracing by Matt Pharr et all (S 97, p 101). Another paper at the frontier of ray-tracing, this work describes strategies to preserve memory coherence (and thus reasonable performace) when ray-tracing massively complex scenes.
• Hierarchical Z-Buffer Visibility by Ned Green, Michael Kass, and Gavin Miller (S93, p 231). This paper presents a really elegant approach to the problem of quickly determining which objects in a scene are visible. One of those "why didn't I think of that?" papers. Someday all interactive graphics may use this technique.
• Hierarchical Image Caching for Accelerated Walkthroughs of Complex Environments by Jonathan Shade, Dani Lischinski, David Salesin, Tony DeRose, and John Snyder (S96, p 75). A nice acceleration technique that performs a spatial subdivision of a scene to be rendered. As the user walks through the scene, nodes at various levels in the spatial hierarchy are rendered and the resulting image is "cached" with that node. The node can then be rendered as a single rectangle texture-mapped with the cached image. The authors also describe an error metric regulating how far the user can move before the original geometry must be re-rendered.
• Computer-Generated Pen-and-Ink Illustration by Georges Winkenbach and David Salesin (S94, p 91). One of the seminal papers in non-photorealistic rendering. The title says it all.
• Reflection from Layered Surfaces due to Subsurface Scattering by Pat Hanrahan and Wolfgang Krueger (S93, p 165). Recognizing the multi-layered nature of surfaces such as skin and leaves enables more realistic modeling. For example, skin can be modeled with three layers: the dermis, the epidermis, and a thin coating of oil.
• Modeling and Rendering Architecture from Photographs: A hybrid geometry- and image-based approach by Paul Debevec, Camillo Taylor, and Jitendra Malik (S96, p 11). Reconstructing a 3-D model from a few photographs is the Holy Grail for much computer vision research. This paper focuses on architectural models, which often have symmetries that simplify the problem, and uses image-based techniques to realistically render interesting scenes without having to build extremely detailed geometric models. The system described in this paper was used to create the film we saw of the Campanile, Berkeley's clock tower. 
• The Lumigraph by Steven Gortler, Radek Grzeszczuk, Richard Szeliski, and Michael Cohen (S96, p 43). One of a pair of important and similar papers in the field of image-based rendering (the other is Light Field Rendering, also presented at SIGGRAPH 96). The lumigraph is a new method of capturing the complete appearance of an object, even objects traditionally very hard to model and render using traditional computer graphics. This paper is somewhat difficult but is an important advance in a very hot new field.
• Painterly Rendering with Curved Brush Strokes of Multiple Sizes by Aaron Hertzmann (S98, p 453). Photoshop and similar programs have long had "oil painting" filters that make an image look like a painting. This paper makes more realistic paintings by more closely mimicing the kinds of brush strokes painters actually use.
• Painterly Rendering For Animation by Barbara Meier (S96, p 477). A new way of rendering 3-D environments with a painted feel. Standard "oil painting" filters take an imput image and make it look painted, but applying the algorithm to (say) a video sequence looks very busy because of the lack of temporal coherence among the brush strokes. In this paper the brush strokes are attached to 3-D objects via a particle system, and thus appear consistent from frame-to-frame. This is the source of the very short "haystack" animation I showed on the first day of class. Could make a really nice project.
• Realistic Modeling and Rendering of Plant Ecosystems by Oliver Deussen et al (S98, p 275). The natural scenes presented in this paper may be the most complex and realistic I have ever seen in computer graphics. The source of the trees-in-a-meadow image I showed on the first day of class. In addition to the modeling and creation of plant ecosystems, this paper presents an interesting approach to managing the incredible complexity of the resulting scenes.
• Interactive Update of Global Illumination Using a Line-Space Hierarchy by George Drettakis and Francois Sillion (S97, p 57). A really elegant hierarchical approach to the radiosity problem that allows truly interactive radiosity - open a door and watch the light stream into the room, bounce around, and correctly illuminate surfaces and shadows. Presenting this paper should probably involve a quick overview of previous radiosity work.
• Metropolis Light Transport by Eric Veach and Leonidas Guibas (S 97, p 65). A nice paper that vastly improves standard Monte Carlo path-tracing (an extension of ray-tracing) quality and time complexity, especially on scenes that are notoriously difficult to render using previous techniques. A good state-of-the-art paper in the field of realistic rendering.
• Modeling and Rendering of Metallic Patinas by Julie Dorsey and Pat Hanrahan (S 96, p 387). A nice treatment of the realistic rendering of aged metallic surfaces.
• Synthetic Topiary by Przemyslaw Prusinkiewicz, Mark James, and Radomir Mech (S94, p 351). Realistic plants, pruned to various shapes, are modeled by environmentally sensitive L-systems, a sort of stochastic grammar for growing branches and leaves. A good example of modeling natural phenomena by simplified simulation—in this case, simulation of plant growing and branching behavior. This paper is a precursor to Realistic Modeling and Rendering of Plant Ecosystems, above.
• The RADIANCE Lighting Simulation and Rendering System by Greg Ward (S94, 459). RADIANCE is UNIX freeware for rendering using Monte Carlo path-tracing methods. This paper provides a good introduction to and overview of modern methods for realistic rendering. It also describes a real system with a large user base, available in source or binary form at http://radsite.lbl.gov/radiance.
• Priority Rendering with a Virtual Reality Address Recalculation Pipeline by Matthew Regan and Ronald Pose (S94, p155). A fascinating paper on how to design your own graphics hardware for virtual reality on a shoestring budget. Two very clever optimizations that foreshadow the image-based rendering watershed.
• IRIS Performer: A High Performance Multiprocessing Toolkit for Real-Time 3D Graphics, by John Rolhf and James Helman (S 94, p 381). Performer is a high-level API for squeezing maximum performance out of high-end SGI systems. We have Performer and we have a high-end SGI, and you might want to use them for your project. This paper also serves as a good overview of the issues, difficulties, and strategies of high-performance rendering.
• Polygon-Assisted JPEG and MPEG Compression of Synthetic Images, by Marc Levoy (S95, p 21). A nice idea for compressing synthetic images for transmission by having a low quality rendering (on the client) and a high quality rendering (on the server). The server transmits the difference between the low- and high-quality renderings. This differential image compresses better than the original high-quality image.
• Motion Warping, by Andrew Witkin and Zoran Popovic (S 95, p 105). How to reuse motion capture data for animation. For example, motion capture of a person walking is warped to include the constraint that they have to step over a block, or duck under a doorway. Or motion capture of a tennis swing could be warped to have the racquet contact the ball at different heights. A different (and simpler) approach to motion capture reuse than the "Spacetime Swing" work we’ve seen by Gleisher.
• Cellular Texture Generation, by Kurt Fleischer et al. (S95, p 239). Textures are "grown" across a surface using a biologically inspired cellular development algorithm. Related to the particle-on-a-surface algorithms we’ve discussed. May have the coolest, strangest SIGGRAPH images ever.
• Hierarchical Polygon Tiling with Coverage Masks, by Ned Greene (S 96, p 65). A follow-up, in some ways, to the Hierarchical Z-Buffer Visibility paper presented by Derek. As before, modifies the standard rendering pipeline in a gee-whiz-why-didn’t-I-think-of-that way.
• Hierarchical View-dependent Structures for Interactive Scene Manipulation, by Normand Briere and Pierre Poulin (S96, p 83). One of the first papers to address interactive ray tracing. Clever data structures, the ray tree and color tree, allow incremental changes to a ray-traced scene viewed by a fixed camera.
• A Perceptually Based Physical Error Metric for Realistic Image Synthesis, by Mahesh Ramasubramanian, Sumanta Pattanaik, and Donald P. Greenberg (S99, 73). A state-of-the-art perceptual rendering technique that uses a detailed model of human visual acuity to avoid wasting computation in a global illumination framework when the result will ultimately be imperceptible.
• LCIS: A Boundary Hierarchy for Detail-Preserving Contrast Reduction, by Jack Tumblin and Greg Turk (S99, 101). Real scenes have luminance levels varying over many orders of magnitude, but our computer screens can only display a couple orders of magnitude. Then again, our eyes can only perceive a couple orders of magnitude at a given time. How can we display high-contrast scenes so as to convey information while accurately representing contrast? Note: Jack Tumblin is a faculty candidate and will be speaking on related work later. This is an excellent chance to evaluate the work of a potential faculty member.
• Modeling and Rendering of Weathered Stone, by Julie Dorsey et al (S99). The title says it all. The application: realistic looking architectural models.
• Fast Computation of Generalized Voronoi Diagrams Using Graphics Hardware, by Kenneth Hoff III et al (S99, 277). A much, much cooler paper than the title would imply. Bouncing pianos and moving face mosaics are only a couple of the cool applications of this very clever computational geometry technique.
• Automatic Image Placement to Provide a Guaranteed Frame Rate, by Daniel G. Aliaga and Anselmo Lastra (S99, p 307). Replace distant objects with textured “backdrops”…good idea. How many backdrop images do we need to precompute and where do they go?
• Teddy: A Sketching Interface for 3D Freeform Design, by Takeo Igarashi (S99, p 409). We’ve all seen Teddy, that cool Java applet for quickly sketching cute rotund genus-0 3-D models. How the hell does it work?
• Art-based Rendering of Fur, Grass, and Trees, by Michael A. Kowalski et al (S99, p 433). The Dr. Seuss paper.
• View-Dependent Geometry, by Paul Rademacher (S99, p 439). Mickey Mouse’s ears are always circular, never elliptical, even under perspective projection. This is physically impossible and downright creepy. How can we make a 3-D model of Mickey?