Program

TexGraph ‘07

May 17

Langford C414

Texas A&M University

Donald House, 2007 Chair

9:50 Opening Remarks

Session I

10:00 Muscle-based Facial Animation Using Blendshapes in Superposition

Andrew Smith, Texas A&M University

Abstract - Blendshapes are an effective tool in computer facial animation, enabling represention of muscle actions. Limitations exist, however, in the level of realism attainable under conventional use of blendshapes as non-intersecting deformations. Using the principle of superposition, it is possible to create a facial model with overlapping blendshapes and achieve more realistic performance. When blendshapes overlap, the region of intersection is in superposition and usually exhibits undesired surface interference. In such cases we use a corrective blendshape to automatically remove the interference. The result is an animatable facial model implemented in Maya which represents the effects of muscle action superposition. Performance created with our model of a known human subject is compared to 3D scan reference data and video reference data of that person. Test animation is compared to video reference footage. The test animation seems to accurately mimic the effects of actual muscle action superposition.

Bio - Andrew Smith is a character artist at Terminal Reality, Inc. He received an M.S. in Visualization Sciences from Texas A&M University in 2006, and a B.S. in Computer Science in 2003, also from Texas A&M.

10:20 Example-based Skeletonization

Can Yuksel, Texas A&M University

Abstract - We present a method for extracting a hierarchical, rigid skeleton from a set of example poses. We then use this skeleton to not only reproduce the example poses, but create new deformations in the same style as the examples. Since rigid skeletons are used by most 3D modeling software, this skeleton and the corresponding vertex weights can be inserted directly into existing production pipelines. To create the skeleton, we first estimate the rigid transformations of the bones using a fast, face clustering approach. We present an efficient method for clustering by providing a Rigid Error Function that finds the best rigid transformation from a set of points in a robust, space efficient manner and supports fast clustering operations. Next, we solve for the vertex weights and enforce locality in the resulting weight distributions. Finally, we use these weights to determine the connectivity and joint locations of the skeleton.

Bio – Can Yuksel earned an MS in Visualization Sciences at Texas A&M University in 2007, and a BS in Physics, with minor in Computer Science and Mathematics at Bogazici University, Istanbul, Turkey in 2004. He was awarded the Pixar-Aggie Scholarship in 2006 for outstanding work on the Viza-Go-Go reel. He will be joining Dreamworks Animation in June 2007 as an Effects Artist.

10:40 Simplifying and Improving Sketch Recognition User Interface Creation

Tracy Hammond, Texas A&M University

Abstract - Sketching is a natural modality for a variety of tasks in human-human interaction and human-computer interaction. Sketch recognition is the process of classifying and understanding hand-shapes to allow for higher reasoning about them later, such as performing simulations, automatic correction, or code generation in the case of software diagrams. Sketch recognition systems have been built for a number of domains, however, they still require significant effort and expertise to build from scratch. We want to enable user interface designers and domain experts to build sketch recognition systems, rather than sketch recognition experts. We also want these sketch recognition systems to be user-independent, allowing sketchers to draw as they would most naturally.

To this end, we have built a framework and system with which sketch recognition systems can be built quickly and without sketch recognition expertise. Designers can build a sketch system by simply sketching a single example of each of the shapes in the domain, and specifying any contextual information to help with recognition.

As part of our over-arching goal, we have worked on the following problems:

- Improving lower-level sketch processing and beautification

- Creating a geometric language (which we have named LADDER) to describe shapes based off of human perception

- Performing user studies to determine how humans naturally describe and perceive shapes

- Determining constraint thresholds from human perception

- Examining HCI issues of inputing shape information

- Automatically generating a shape description from a single drawn example using perceptual cues.

- Generating sample shape from a single description

- Generating near miss examples to clarify the example

- Creating a novel form of active learning in which the system generates its own (near-miss) examples, and uses the teacher as a source of labels.

- Creating a novel form of the version spaces algorithm to handle interrelated constraints and has the ability to also learn negative and disjunctive constraints

Bio - Tracy Hammond is an assistant professor at Texas A&M University with a focus on human perception, sketch recognition, computer human interaction, and learning. She earned the B.A. in math, the B.S. in applied math, the M.S. in computer science, the M.A. in anthropology from Columbia University and the PhD in computer science from MIT. Previously, she taught for five years at Columbia University, and she was a telecom analyst for four years at Goldman Sachs, where she designed, developed, implemented, and administered global computer telephony applications.

11:10 Interactive Creation and Editing of 3D Deformations for Human Lung Modeling

Travis McPhail, Rice University

Abstract – Computed Tomography (CT) scans are one of the major tools used in lung cancer treatment planning. Individuals use CT scans to extract lung models and compute deformations of patients’ lungs during their breath cycle. From these deformations, doctors attempt to measure quantities such as ventilation (the rate at which air enters/leaves lungs) and perfusion (the flow of blood in lungs). There have been a number of attempts to automatically compute deformations between successive ct scans, however due to the considerable amount of noise in these images, human-assisted heuristic methods are needed. We present an interactive tool for creating point-based deformations. This tool semi-automatically finds point correspondences between successive images and uses the GPU to locally view the quality of these point correspondences.

Bio – Travis McPhail is a graduate student in the Computer Science Department of Rice University in Houston, TX. He received his B.S. and M.S. at Rice University and is continuing to his Ph.D. His research interests are in biomedical imaging, deformable modeling, and computer game development. His primary focus is biomedical imaging.

11:30 Invited Talk - The Devil is in the Details

Geoff Wyvill – University of Otago, New Zealand

Abstract - Programming is a creative activity and, so far, no one has been able to provide a formula for doing it well just as no one has a formula for writing a best selling novel. In the last forty years we have seen 'methodologies', 'structured programming', 'object oriented programming', 'reusable code' and any number of other doctrines. Meanwhile programs have got bigger and done less. I examine just a few examples of wasteful coding, particularly in graphics. By studying these small cases we can get a handle on the art of writing cleaner, faster and smaller code.

Bio - Professor Geoff Wyvill has a BA in Physics from Oxford and MSc and PhD in Computer Science from Bradford University. His research interests includes modelling, csg, antialiasing, ray tracing, surface design, implicit methods and the application of computers in the arts. His teaching over 30 years has included programming, data structure, theory of computing, complexity, compilers, various languages, computer graphics, history of computing, visualization methods. He is in editorial boards of The Visual Computer, Computer Animation, International and Virtual Worlds, Journal of Shape Modeling and Computer Graphics Forum.

12:00 Lunch

Session II

1:30 Online Motion Capture Marker Labeling for Multiple Interacting Articulated Targets

Qing Li, University of Houston

Abstract - In this paper, we propose an online motion capture marker labeling approach for multiple articulated interacting targets. Given hundreds (or even thousands) of un-labeled motion capture markers from multiple captured targets that are interacting each other, our approach can automatically label these markers frame by frame, by fitting rigid bodies and exploiting trained structure and motion models. The advantages of our approach include: 1) our method is an online algorithm, which requires no user interaction once the algorithm starts. 2) Our method is much more robust than the previous the closest point-based approaches by automatically imposing the structure and motion models. 3) Due to the use of the structure model which encodes the rigidity of each articulated body of captured targets, our method can recover the missing markers reliably. Our approach is efficient and is particularly suited for online computer animation and video game applications.

Bio – Qing Li is a Ph.D. student in Computer Science at the University of Houston.

1:50 Computer Graphics at TAMU-CC

Scott King, Texas A&M University - Corpus Christi

Abstract - Texas A&M University - Corpus Christi is a growing regional institution. We offer a BS and MS in computer science with research interests in networking (security, wireless, sensor nets), HCI, Databases, Parallel computing and graphics. This talk will discuss our efforts in teaching and research in computer graphics.

Bio - Scott King is an Assistant Professor in the Computing Sciences Department at Texas A&M University-Corpus Christi. He received his M.S. and Ph.D from The Ohio State University. His research interests are in computer graphics and parallel computing. His primary focus is on animation and real-time graphics.

2:20 Line Drawing as a Dynamic Process

Mayank Singh, Texas A&M University

Abstract – We introduce a novel mechanism for creating silhouette line drawings from polygonal models that captures the dynamic nature of the drawing process. The approach takes into account the interaction between the moving human hand and the drawing instrument. The pen is treated as a physically-based object with mass and thus momentum. A control system drives the pen by tracking the contour of the polygonal model as projected onto the drawing surface, thus mimicking hand motion. The generated lines have a smooth hand-drawn quality. Lines are rendered using a ribbon metaphor, where thickness is determined by the twist of the ribbon. The twist angle can be dependent upon various attributes such as perspective depth, the curvature of the line, and the lighting of the model. A number of examples are presented, ranging from tightly controlled drawings to expressive gestural drawings.

Bio – Mayank Singh received his Bachelors in Architecture from IIT-Roorkee, India in 2000 and his Masters in Computer Science from Marquette University in 2002. He is currently a Ph.D. student with the Department of Computer Science working with the Visualization Lab. His research interests include Non Photorealistic Techniques, Illustrative & Scientific Visualization.

2:40 A Virtual Sculpture Based Morphable Face Model

Jessica Riewe, Texas A&M University

Abstract - Exemplar virtual three-dimensional sculptures collected from sixteen artists are used to develop a multi-dimensional space to describe facial shape, which is then used as a mechanism for modeling new faces. This is accomplished through identifying and varying derived principal components of the multi-dimensional space. The relationships between these principal components and their effects on new faces are explored through the creation of new virtual faces using a graphical user interface. These new virtual face sculptures are then modified by facial feature, gender, and expression using a feature-based transformation interface based on difference vectors. Finally, animations are created to illustrate the results of these approaches. Facial mesh transformations based on principal components does not give direct predictable control for modifying specific facial features. However, it does provide interesting design choices for artists, and general patterns of the data variance were obtained. The feature-based transformations were successful in further modifying created faces.

Bio - Jessica Riewe is a recent graduate from the Texas A&M Visualization Laboratory. She completed her undergraduate at A&M in Environmental Design.

3:00 Break

Session III

3:20 Mu-bases and Their Applications in Geometric Modeling

Ning Song, Rice University

Abstract - Geometric modeling is the sub-field of computer science concerned with constructing, manipulating, and analyzing geometric models – computer models of physical, virtual or mathematical objects. In practice, many geometric models are represented by smooth algebraic functions, especially rational functions. For rational curves and surfaces, there are several common problems that are frequently encountered: computing intersections, finding implicit equations, and detecting singular points and base points. Generally, each of these problems eventually reduces to solving systems of polynomial equations.

This talk will focus on solving these common problems by mu-bases. The notion of a mu-basis is defined for an arbitrary number of polynomials in one variable. The properties of these mu-bases are derived, and a straightforward algorithm is provided to calculate a mu-basis for any collection of univariate polynomials. Systems where base points are present are also discussed. mu-bases are then applied to solve implicitization, inversion and intersection problems for rational space curves.

Bio – Ning Song is a PhD Candidate in the Computer Science Department at Rice University. He holds an M.S. in Computer Science from Rice, and a B.S. in Computer Science from the University of Science and Technology, China.

3:40 Wave Particles

Cem Yuksel, Texas A&M University

Abstract – We present a new method for the real-time simulation of fluid surface waves and their interactions with floating objects. The method is based on the new concept of wave particles, which offers a simple, fast, and unconditionally stable approach to wave simulation. We show how graphics hardware can be used to convert wave particles to a height-field surface, which is warped horizontally to account for local wave-induced flow. The method is appropriate for most fluid simulation situations that do not involve significant global flow. It is demonstrated to work well in constrained areas, including wave reflections off of boundaries, and in unconstrained areas, such as an ocean surface. Interactions with floating objects are easily integrated by including wave forces on the objects and wave generation due to object motion. Theoretical foundations and implementation details are provided, and experiments demonstrate that we achieve plausible realism. Timing studies show that the method is scalable to allow simulation of wave interaction with several hundreds of objects at real-time rates.

Bio – Cem Yuksel is a PhD student in the Department of Computer Science at Texas A&M University. He received his B.S. in Physics and M.S. in Computer Engineering from Bogazici University in Istanbul, Turkey. His research interests are physically based simulations, hair simulation, real-time and offline rendering algorithms, and global illumination. Cem’s Wave Particle work will be featured in the paper and animation theater venues at Siggraph 2007.

4:00 Visualization of Prediction Model Errors

Larry Young, Texas A&M University - Corpus Christi

Abstract - One of the common problems of research scientists is that predicted results often differ from measured results. This is known as prediction error. This project developed a system to visualize prediction error called Visualization of Model Prediction Errors or VMPE. VMPE was designed as a tool to allow modelers to see discrepancies between model outputs and what actually occurred, and to concurrently display multiple related data series, typically inputs to the model.

The VMPE system is point and time based, in that all data are expected to be tagged with a location and a time. Data for multiple points and different time periods can be overlaid via a Windows graphical user interface. The user may select a specific time period for more intense scrutiny, as well as selecting which data to include in the visualization. This project was designed to fill a niche, visualizing prediction errors, and provide something of substance and usefulness to researchers who are developing predictive models that can be compared to actual results. Model builders will benefit by having a visualization system specifically designed to visualize prediction error and correlate those errors with the data input to the model.

Bio - Larry Young is currently an Instructor of Computer Science at Texas A&M University-Corpus Christi. He completed a Masters Degree in Computer Science from Texas A&M University-Corpus Christi in 2007 and received a Masters of Business Administration from Golden Gate University in 1984. He retired as a Lt Col in the United States Air Force after 22 years of service as a Communications/Computer Office.

4:20 A Grid with a View: Optimal Texturing for Perception of Layered Surface Shape

Alethea Bair, Texas A&M University

Abstract – We present the results of two controlled studies comparing the ability to estimate surface normal for layered surface visualizations under various surface texture conditions. Error of a hand-set surface normal probe was used for comparison. The no-texture, layered-surface case was compared both with a single surface case and layered surfaces with projected grid textures. Variations in relative texture size on top and bottom surfaces were compared, as well as opacity of the top surface. Significant improvements are found for the textured cases over non-textured surfaces. Either larger or thinner top-surface textures, and lower top surface opacities are shown to give less bottom surface error. Top surface error appears to be highly resilient to changes in texture. Given the results, we also present an example of how appropriate textures might be useful in volume visualization

Bio – Alethea Bair is a Ph.D. student at Texas A&M University in the Department of Architecture, where she is pursuing a program of study in Visualization Sciences. She received a degree in physics from the University of Illinois, Urbana/Champaign, and as an undergraduate did summer research applying visualization techniques in Physics at the University of Wisconsin. Her research interests include perceptual optimization, data mining and pattern analysis, and image analysis.

4:40 Really Hot Stuff: Driving Object Deformations from Internal Physical Processes

Zeki Melek, Texas A&M University

Abstract - We present a method for deforming objects for graphics applications, based on the results of internal physical simulations. As driving examples, we describe in detail methods for simulating the bending of burning matches, and the crumpling of burning paper. In these cases, the small-scale changes in a chemical process result in large-scale deformations of the given object. We propose the use of a free form deformation to model such large-scale deformations.

Bio - Zeki Melek received MS and BS degrees in Computer Science from Bogazici University, Turkey in 96 and 00 respectively. He has been with the Department of Computer Science at Texas A&M University since 2001, working with Dr. Keyser. His dissertation title is "Interactive Simulation of Fire, Burn and Decomposition".

5:00 Closing Remarks