3D Scene Understanding – It's Time to Address it Again
Inspired by the ability of humans to interpret and understand 3D scenes nearly effortlessly, the problem of 3D scene understanding has long been advocated as the ?holy grail? of computer vision. In the early days this problem was addressed in a bottom-up fashion without enabling satisfactory or reliable results for scenes of realistic complexity. In recent years there has been considerable progress on many sub-problems of the overall 3D scene understanding problem. As the performance for these sub-tasks starts to achieve remarkable performance levels we believe that the problem to automatically infer and understand 3D scenes should be addressed again. In this talk we will ? on the one hand ? highlight progress on some essential components of scene understanding such as object class recognition and articulated pose estimation and tracking. On the other hand we will also report on our current attempt towards 3D scene understanding in the particular case of traffic scene analysis.
Bernt Schiele is Max-Planck-Director at MPI Informatics and Professor at Saarland University since 2010.
He studied computer science at the University of Karlsruhe, Germany. He worked on his master thesis in the field of robotics in Grenoble, France, where he also obtained the “diplome d'etudes approfondies d'informa-tique”. In 1994 he worked in the field of multi-modal human-computer interfaces at Carnegie Mellon Universi-ty, Pittsburgh, PA, USA in the group of Alex Waibel. In 1997 he obtained his PhD from INP Grenoble, France under the supervision of Prof. James L. Crowley in the field of computer vision. The title of his thesis was “Ob-ject Recognition using Multidimensional Receptive Field Histograms”. Between 1997 and 2000 he was postdoctoral associate and Visiting Assistant Professor with the group of Prof. Alex Pentland at the Media Laboratory of the Massachusetts Institute of Technology, Cambridge, MA, USA. From 1999 until 2004 he was Assistant Professor at the Swiss Federal Institute of Technology in Zurich (ETH Zurich). Between 2004 and 2010 he was Full Professor at the computer science department of TU Darmstadt.
Geometric Modeling on Different Levels of Abstraction
For a number of years, performance and surface smoothness have been the driving forces in the development of new geometry processing algorithms. Structure and shape analysis have mostly been considered as an independent pre-processing stage in order to partition a given geometric model into segments. This segmentation was then used to restrict the scope of certain mesh operations such as parametrization for texturing or reverse engineering, deformation, and re-meshing. The recent trend towards a higher and higher complexity of 3D models and the diversification of geometric modeling and processing into a much wider range of application domains (CAD/CAM, Simulation, Visualization, Games, Animation, Medicine, ?) has revealed the need to introduce higher levels of abstraction into geometric models in order to manage this versatility. At the same time, new (interactive) modeling paradigms are investigated which facilitate the exploration of shape design spaces through intuitive interfaces. Hence, on the one hand an integrated view on model representation and shape control is desirable for the sake of efficiency but on the other hand, the diversity of requirements in different application scenarios require that shape control has to be made independent from the underlying representation. This is achieved by introducing different levels of abstraction for geometric models.
In my talk, I propose a taxonomy which distinguishes (1) the representation layer, (2) the feature layer, (3) the control layer and (4) the constraint layer. I will present a number of recent projects performed by the Comput-er Graphics Group at RWTH Aachen University which generate and exploit these levels of abstraction in order to solve specific geometric modeling tasks. It will include examples for model augmentation, i.e. the transfor-mation of raw triangle meshes into meshes whose structure better captures the global features of the underly-ing geometric shape by proper orientation and alignment of polygonal faces (quad mesh/layout generation). Moreover, I will present shape modeling techniques that abstract from the particular representation and allow for effective and intuitive interactive shape modifications. Through the introduction of (potentially non-linear) constraints, we can further abstract from the underlying representation by restricting shape modifications to ?plausible? or ?admissible? designs.
Leif Kobbelt graduated with a PhD in Computer Science at the University of Karlsruhe in 1994. After a postdoc stay at the University of Wisconsin he joined the Computer Graphics Group at the University of Erlangen in 1996 where he completed his Habilitation in 1999. Shortly after being appointed an associate professor at MPI Informatik in Saarbr?cken, he received an offer for a full professorship from RWTH Aachen University and moved to Aachen in 2001 where he is the head of the Computer Graphics and Multimedia Group. His research interests include most areas of Computer Graphics and Computer Vision with a focus on Geometry Processing, Multiresolution and Freeform Modeling, 3D Model Optimization, and the efficient handling of polygonal meshes. He is involved in diverse research collaborations in application fields ranging from CAD/CAM and medical image processing to mobile multimedia applications, simulation sciences, and psychology. He cooperates with research groups and companies in Europe, North America and Asia. Leif's research work during the last years resulted in numerous widely cited publications in top scientific journals and international conferences. For his scientific contributions he received several renowned awards, including the Heinz-Maier-Leibnitz Award 2000 from the German Federal Government and the Outstanding Technical Contribution Award 2004 from the Eurographics Association. In 2008 he was named a Eurographics fellow. He frequently serves on program committees of major international conferences and organized and chaired several workshops and conferences. He is a reviewer and advisor for a number of national and international research agencies and consults several industry companies.
Visual tools for understanding multi-dimensional parameter spaces
Simulations are an integral part of computational science. Simulations are characterized by a particular set of inputs and a multitude of outputs. Understanding the dependency of the outputs from the inputs is key for understanding the underlying phenomena that are being modeled. In this talk I will try to give a characteriza-tion of such general input/output systems and present several tools we have built for different applications, ranging from fisheries science to medical imaging to fluid simulation. I will try to make the case that visual support greatly facilitates the understanding of these complex systems.
Torsten M?ller is a professor at the School of Computing Science at Simon Fraser University. He received his PhD in Computer and Information Science from Ohio State University in 1999 and a Vordiplom (BSc) in mathe-matical computer science from Humboldt University of Berlin, Germany. He is a senior member of IEEE and ACM, and a member of Eurographics. His research interests include the fields of Visualization and Computer Graphics, especially the mathematical foundations thereof.
He is co-director of the Graphics, Usability and Visualization Lab (GrUVi). He is the appointed Vice Chair for Publications of the IEEE Visualization and Graphics Technical Committee (VGTC). He has served on a number of program committees and has been papers co-chair for IEEE Visualization, EuroVis, Graphics Interface, and the Workshop on Volume Graphics as well as the Visualization track of the 2007 International Symposium on Visual Computing. He has also co-organized the 2004 Workshop on Mathematical Foundations of Scientific Visualization, Computer Graphics, and Massive Data Exploration as well as the 2010 Workshop on Sampling and Reconstruction: Applications and Advances at the Banff International Research Station, Canada. He is a co-founding chair of the Symposium on Biological Data Visualization (BioVis). In 2010, he was the recipient of the NSERC DAS award.