学术讲座题目：Nonrigid Graphics: Principles and Practices 主讲人：新加坡南洋理工大学教授 Lin Feng 时间：2013年1月11日 上午9点 闵行校区技术中心3B306会议室 欢迎广大师生参加。 Abstract:
We introduce a theoretic framework of nonrigid graphics for modeling and rendering of deformable graphic objects at thecore graphics system layer. Furthermore, we aim to implement aninnovative core graphics system in the form of kernels directly on the new-generation Graphics Processing Unit (GPU). Onone hand, the most challenging tasks in next-generation cyberspaceapplications are to model and render objects and phenomena withcomplex properties such as nonuniform and nonrigid materials,growing and differentiating structures, collision-aware andhaptic-sensitive surfaces, etc. On the other hand, current computergraphics has been based on rigid geometric models, in which agraphic scene is decomposed into rigid primitives such as triangleand tetrahedron in the core graphics system; modeling and viewingthese primitives are controlled by affine transforms. Applicationsinvolving nonrigid transformation require the programmers todevelop the application data-specific deformation modules, beingphysically-based or not, at the application layer. Thesetransformation modules are then unavoidably translated into a rigidgraphics implementation system via a standard API such as OpenGLTM and Direct3DTM . Thus, even though a mathematical model may exist for non-affinetransformation at the application layer, it usually employs acomputationally expensive numerical solution which has to beimplemented in the high-level application modules.Trans-representation of the deformable models by the rigidprimitives disables direct access to the accelerating architectureimplemented in the low-level core graphics system. Real-timeinteractions with these complex models are impossible. In response,we propose a fundamentally different approach, nonrigid graphicskernels for GPU, which addresses the capability of directrepresentation of nonrigid objects and physically-baseddeformation, specifically, (a) a new theoretical framework ofnonrigid graphics in which a unique tetrahedral mass-springrepresentation will be invented as the “soft object” primitivedirectly accessible by the Shaders in GPU; (b) a physically-basednumerical deformation model which characterizes the elasticity andviscoelasticity of objects in geometric transformation, and thenumerical processes are incorporated into the real-time renderingpipeline for visibility; and (c) a transform feedback architectureand numerical process kernels on GPU for a novel hardwareacceleration solution to modeling and rendering of the deformationprocesses, leveraging on the new-generation GPU technologies,especially the Vertex Shader, Tessellation and Geometric Shader.The significance of the proposed project is twofold: (a) It has afundamental scientific impact on the disciplinary advancement ofcomputer graphics. The proposed theoretic framework will pave theway to direct representations and operations on deformable objectsand nonrigid transformation in future core graphics systems. (b)Furthermore, it has a wide application domain in digital mediaindustries. Real-time applications such as 3D image registration incomputer-aided surgery, scientific simulation and visualization,computer animation, interactive games and virtual reality rely onnot only accurate physically-based modeling but also real-timerendering of the deformation processes. The GPU kernels to bedeveloped in this project will make them possible.
Speaker:
Dr Lin Feng is an Associate Professor of the Schoolof Computer Engineering, Director of Bioinformatics ResearchCentre, and Programme Director of MSc in Digital Media Technology,Nanyang Technological University, Singapore. He obtained his PhD inComputer Engineering in 1996. He won 7 prestigious research awards,including the “National Science and Technology Advance Award(Second Class)” by Chinese State Council in 1992. He serves asAssociate Editor / Editorial Board for 5 journals including Journal of Signal Processing Systems, Springer , and the Guest Editor of Special Issues on Computing Architectures and Acceleration for BioinformaticsAlgorithms and Biomedical Imaging . He published more than 170 technical papers. He is a SeniorMember of IEEE.
|
