Visual Computing

Institute of Computer Science — Department II

The Department of Visual Computing at the University of Bonn is dedicated to various aspects of visual information, including 2D/3D images and shapes, volumetric videos, 3D objects and scenes. This data is combined with modern methods of data analysis and visualization, augmented and virtual reality, telepresence and/or visual human-machine interaction. They form the basis for a wide range of current and future applications. These include (bio)medical analysis and visualization, virtual experiments in the basic sciences, virtual production, 3D games and the metaverse.

In our research, we pursue a comprehensive, interdisciplinary approach that integrates and combines a spectrum of different scientific disciplines and employs cross-disciplinary research methods. Our efforts represent a mix of methods from fields such as computer vision, sensory data processing, geometric computation, generative modeling, pattern recognition, physically informed simulation, rendering, and visual analysis. This integration is based on a solid theoretical framework of optimization strategies complemented by the application of modern Machine Learning and Deep Learning methods. Our goal is to improve the synthesis, processing, and analysis of visual data, increase efficiency and deepen the understanding of complex visual data structures.

Research Highlights

PDE Approximation: How Hermite Spline Networks Solve Physical Problems Quickly

Our Spline-PINN approach allows fast simulations of dynamic Partial Differential Equations (PDEs), e.g. fluids. For this purpose, neural networks are trained physically and self-monitored.

To the publication

Machine Learning for 3D Shape Analysis

Our 3D shape matching solution uses Deep Learning with spectral regularization in unsupervised training. It outperforms previous approaches in matching quality in nine data sets.

To the publication

Hair Rendering: Our Microfacet Model Sets New Standards in Computer Graphics

We present new physically plausible as well as efficiently calculable models for light scattering from feathers and hair.

To the publication

Efficient Algorithms for Precise Fiber Tract Reconstruction in the Brain

We develop robust and efficient algorithms to reconstruct fiber tracts in the human brain. The results are used in surgical planning and neuroscience.

To the publication

Working Groups

Department II "Visual Computing" consists of several working groups under the direction of Prof. Dr. Florian Bernard, Prof. Dr. Matthias B. Hullin, Prof. Dr. Reinhard Klein, Prof. Dr. Thomas Schultz and Dr. Eduard Zell.

The research group "Learning and Optimization for Visual Computing" strives for intelligent visual data processing by combining computational models with advanced Machine Learning techniques to incorporate human knowledge. This prevents machine learning systems from having to expend unnecessary resources to relearn what is already known and offers a wide range of potential applications in areas such as autonomous driving, medical image analysis, human performance capture and video and film production.

The "Digital Material Appearance" working group is concerned with measuring, simulating, analyzing and inverting the propagation of light in complex optical constellations. The models and processes developed in the working group not only enable more convincing digital twins of real-world objects, but also provide new analysis tools for the geometry and optical properties of the world around us.

In the "Computer Graphics" working group, algorithms and methods in the fields of geometry processing, 3D vision, rendering, simulation and real-time graphics are developed on a profound theoretical basis, equipped with solid knowledge of Machine Learning. The motivation provides open problems in applications such as 3D game development, medical image processing, architecture, engineering and special challenges such as the development of a telepresence system.

Evaluating the large, complex and dynamic amounts of image data generated by research and medicine requires support from automatic image analysis as well as interactive visualization, Machine Learning and mathematical modelling. The "Visualization and Medical Image Analysis" working group's research in this area ranges from methodological foundations to numerous collaborations with users within the University of Bonn and beyond.

Dr. Zorah Lähner's “Geometry in Machine Learning” working group focuses on the analysis and optimization of geometric objects using machine learning approaches. Geometric optimization has the potential to improve algorithms and neural networks in many ways, e.g. by reducing the dimensionality of problems, simplifying representation or using approximation techniques. Possible areas of application are in the fields of virtual reality, medicine, and physics.