Opening of Intel Visual Computing Institute
Scientific
Program of Saarland University and
Research Institutes
Opening, may 12th, 2009, 10.00 to 11.50
Press Conference, 11.50 to 12.20
Scientific Program, 12.20 to 13:00
Aula of Saarland University
University Campus, Building A 3.3
Hybrid
Error Correction
Thorsten
Herfet,
Professor of Telecommunications
Audio media emerge to be the dominating content delivered within the Future Media Internet. They impose new challenges to protocols and coding schemes:Timely delivery as well as optimal use of the available bandwidth are essential.The demonstrated Hybrid Error Correction scheme operates at near- optimal efficiency and provides partial reliability adapted to the individual demands of the visual application.
Markerless Performance Capture, Virtual Actors and 3D Video Processing
Carsten Stoll, PhD-Student at Max Planck Institute for Computer Science, Computer Graphic Department of Professor Hans-Peter Seidel
This demo presents algorithms developed for capturing and processing human performances from multi-view video. We reconstruct spatio-temporally coherent geometry, motion and textural surface appearance of actors, even when wearing wide apparel or skirts.
Camera Motion Estimation and 3D Reconstruction from Video Abstract
Thorsten Thormählen, Head of the independent research group within the Max Planck Center for Visual Computing and Communication
Thorsten Thormählen presents tools and algorithms for image-based 3D scene analysis. This comprises the estimation of camera motion of one or more cameras in a given scene, interactive techniques for the reconstruction of the static scene geometry, and algorithms for motion and shape estimation of moving objects.
Statistical Geometry Processing and Geometric Data Analysis
Michael
Wand,
Head of the independent research group within the Cluster
of Excellence "Multi-Modal Computing and Interaction"
The
demo of Michael Wand consists of two parts: First, we demonstrate a
realtime rendering and editing system for very large point clouds
from 3D scanners. It is currently the only available system for
interactive editing of extremely large geometric data sets. Secondly,
we show symmetry detection techniques that fully automatically
decompose 3D scans into building blocks (for example the windows of a
facade scan). This information can be used subsequently for various
applications, such as noise removal and hole filling.
Dense 3D Reconstruction Using Optical Flow
Research Group “Mathematical Image Analysis Group” at Saarland University, Levi Valgaerts, Andres Bruhn, Markus Mainberger, Pascal Gwosdek and Professor Joachim Weickert
This research group demonstrate how state of the art computer vision techniques can be used for 3D reconstruction of buildings and human faces from stereo image pairs. Our application determines the optical flow between two input images and subsequently computes their geometric relation in the form of the fundamental matrix. By using only internal camera calibration information, such as the focal length, we can perform a triangulation in 3D space. The result is a dense surface mesh of the depicted scene.
Edge-Based Image Compression
Markus
Mainberger and Professor
Joachim Weickert, Research Group “Mathematical Image Analysis
Group” at Saarland University
We present a system where digital images can be encoded in an efficient way by storing only the colours on both sides of the most important edges. For reconstructing the missing information, inpainting with the Laplace equation is performed. Since this can be achieved with highly efficient multigrid algorithms, the codec can be used in real-time applications. This is an example of a novel class of lossy image compression methods that are based on partial differential equations (PDEs). They can built the basis for future alternatives to standards such as JPEG or JPEG 2000.
Distributed Ray-Tracing and Synchronous Distributed Output
Research Group of Professor Philipp Slusallek, Computer Graphics at Saarland University and German Research Center for Artificial Intelligence (DFKI)
Our framework for distributed rendering and display is based on the Network-Integrated Multimedia Middleware (NMM). It allows to distribute rendering tasks on multiple CPUs and GPUs of a local machine as well as among several machines in a network. The framework takes care of load balancing and synchronization of this "render graph" between the different processing units.
Ballview: Advanced Rendering Techniques Applied to Biological Molecules
BALLView is an interactive viewer for molecule data. It helps researchers to perceive the complex structure of biological molecules. The knowledge of such structures is crucial when developing new agents in drug design. Advanced visualization methods support the understanding of molecule configurations and the interactions among them.
RTfact: Ray-Tracing Core Development Architecture,
Professor Philipp Slusallek, Ilyan Georgiev and others
Future architectures of processing units allow for the massively parallel execution of tasks. Since the Ray-Tracing algorithm is inherently parallel, we expect huge benefits and an increase in Ray-Tracing performance. Nevertheless new problems arise from this task parallelism. High numbers of different tasks accessing different data in memory can have a disastrous impact on system performance. The Ray-Tracing core development architecture RTfact solves those problems and yields a flexible platform for rendering research.
Architectural Planning using Modern Visualization
Computer Graphics Group of Professor Philipp Slusallek, University and DFKI
Computer Aided Design takes a leap from pure engineering applications to the field of architecture. New modeling tools enable architects to design buildings in 3D, plan heating, electrical and sanitary installations and run simulations with these models. The complexity of these models will grow in the near future, as it has been the case in automobile and aircraft industry. But other than in these fields buildings are closely linked to their environment. We show how data from different sources like architects, airborne laser scanning, digital ariel photography and other geo referenced data can be combined into a single model, which allows to assess the architectural design and to judge the influence of a building to its environment.
Scientific
Visualization of Molecular Volume Data Sets
Group of Professor Philipp Slusallek
X-Ray Crystallography is the main tool to capture the structure of molecules. Biological molecules are turned into crystals, which are then analyzed with X-Rays. This does not yield the positions of the atoms comprising the molecule, but an electron density distribution within the molecule. Since turning molecules especially proteins into crystals is very hard, the measurements of the electron density are prone to errors. Our Direct Volume Renderer allows to easily visualize the distribution of electrons within the molecules and thus is a good tool to determine the possible structure of molecules.