Dr. rer. nat. Steffen Frey
Visualization Research Institute, University of Stuttgart

Room 00.110
Allmandring 19
70569 Stuttgart
Tel.: +49 (0)711 685-88629
Stacks Image 281
The purpose of scientific visualization is to graphically illustrate scientific data, enabling scientists to gain new insights and a deeper understanding. The size and complexity of these data sets are continuously increasing with new simulation and data acquisition methods as well as growing hardware processing power. Visualization needs to be able scale with the data, and therefore itself needs to develop new illustration techniques in conjunction with novel approaches harness the huge processing power of parallel environments.

To accomplish this goal, I develop new approaches for the efficient usage of parallel environments (in particular GPUs and clusters) as well as the illustration of the data, with the main goal being the integration of both aspects. While the emphasis is more on one or the other depending on the specific objective, neither component may be completely neglected to achieve both expressive and responsive visualization.
Stacks Image 1040
Extraction of Fragments and Waves After Impact Damage in Particle-Based Simulations
P. Diehl, M. Bußler, D. Pflüger, S. Frey, T. Ertl, F. Sadlo, M. A. Schweitzer
Meshfree Methods for Partial Differential Equations VIII, Springer International Publishing
The analysis of simulation results with complex geometries and the verification against experimental data is essential for impact damage and wave propagation. We present two visualization techniques for post-processing particle-based simulation data and highlight new aspects for the quantitative comparison with experimental data. Peridynamics, a non-local generalization of continuum mechanics, is considered as the reference particle method. The first approach is an extended connected component algorithm to extract the fragment size and the corresponding histograms. The distribution of the fragment size is experimental verifiable with these experiments. The second approach focus on the visualization of the stress after an impact in a complex geometry. Here the particle-based data is re-sampled and rendered with standard volume rendering techniques to address the interference pat- tern of the stress wave after the reflection at the boundary.
Stacks Image 1186
Balanced Sampling and Compression for Remote Visualization
S. Frey, F. Sadlo, and T. Ertl
We present a novel approach for handling sampling and compression in remote visualization in an integrative fashion. As adaptive sampling and compression share the same underlying concepts and criteria, the times spent for visualization and transfer can be balanced directly to optimize the image quality that can be achieved within a prescribed time window. Our dynamic adjustments regarding adaptive sampling, compression, and balancing, employ regression analysis-based error estimation which is carried out individually for each image block of a visualization frame. Our approach is tuned for high parallel efficiency in GPU-based remote visualization. We demonstrate its utility within a prototypical remote volume visualization pipeline by means of different datasets and configurations.
Stacks Image 1105
Exploratory Performance Analysis and Tuning of Parallel Interactive Volume Visualization on Large Displays
A. Panagiotidis, S. Frey, and T. Ertl
EuroVis 2015, Short Paper, Cagliari, Sardinia, Italy, 2015.
In this paper, we present an integrated approach for exploratory performance analysis and parameter optimization of interactive distributed volume visualization for large displays. We collect performance metrics of interest on-the-fly directly from both the GPU and our volume ray casting implementation and visualize them simultaneously. This allows users to explore the data set together with the corresponding metrics to jointly investigate both the visual and the performance impact of different parameter settings, like camera position, sampling density, or acceleration technique.
Stacks Image 1031
On In-Situ Visualization for Strongly Coupled Partitioned Fluid-Structure Interaction
O. Fernandes, D.. Blom, S. Frey, A. V. Zuijlen, H. Bijl, and T. Ertl
We present an integrated in-situ visualization approach for partitioned multi-physics simulation of fluid-structure interaction. The simulation itself is treated as a black box and only the information at the fluid-structure interface is considered, and communicated between the fluid and solid solvers with a separate coupling tool.
Stacks Image 1487
Interactive Progressive Visualization with Space-Time Error Control
S. Frey, F. Sadlo, K.-L. Ma, and T. Ertl
Static settings with respect to a certain image quality or frame rate are inherently incapable of delivering both high frame rate for rapid changes and high image quality for detailed investigation. Our technique flexibly adapts by steering the visualization process in three major degrees of freedom: when to terminate the
refinement of a frame in the background and start a new one, when to display a frame currently computed, and how much resources to consume. We base these decisions on the correlation of the errors due to insufficient sampling and response delay, which we estimate separately using fast yet expressive heuristics.
Stacks Image 132
Space-Time Volumetric Depth Images for In-Situ Visualization
O. Fernandes, S. Frey, F. Sadlo, and T. Ertl
Volumetric depth images (VDI) are a view-dependent representation that combines the high quality of images with the explorability of 3D fields. By compressing the scalar data along view rays into sets of coherent supersegments, VDIs provide an efficient representation that supports a-posteriori changes of camera parameters. In this paper, we introduce space-time VDIs that achieve the data reduction that is required for efficient in-situ
visualization, while still maintaining spatiotemporal flexibility.
Stacks Image 1570
Explorable Volumetric Depth Images from Raycasting
S. Frey, F. Sadlo, and T. Ertl
View-dependent image-based rendering techniques have become increasingly popular as they combine the high quality of images with the explorability of interactive techniques. However, in the context of volume rendering, previous approaches suffer from various shortcomings, including the limitation to surfaces, expensive generation, and insufficient occlusion and motion parallax impairing depth perception. In this paper, we propose Volumetric Depth Images (VDI) to overcome these issues for view-dependent volume visualization by an extension of the Layered Depth Image (LDI) approach. Instead of only saving for each view ray of one camera configuration the depth and color values for a set of surfaces, as in LDIs, VDIs store so- called supersegments, each consisting of a depth range as well as composited color and opacity.
Stacks Image 1653
Mesh Generation From Layered Depth Images Using Isosurface Raycasting
S. Frey, F. Sadlo, and T. Ertl
This paper presents an approach for the fast generation of meshes from Layered Depth Images (LDI), a representation that is inde- pendent of the underlying data structure and widely used in image-based rendering. LDIs can be quickly generated from high-quality, yet compu- tationally expensive isosurface raycasters that are available for a wide range of different types of data. We propose a fast technique to extract meshes from one or several LDIs which can then be rendered for fast, yet high-quality analysis with comparatively low hardware requirements. To further improve quality, we also investigate mesh geometry merging and adaptive refinement, both for triangle and quad meshes.
Stacks Image 1079
Parallel Interactive Visualization: Strategies and Examples
S. Frey and T. Ertl
Interaction and HPC: Multi-Scale / Multi-Physics Applications, ParCo 2013, Munich, Germany, 2013. (Talk Only)
Scientific visualization applications typically exhibit a common structure and share a number of characteristic properties and requirements in the context of parallel computation. We outline these aspects by means of several applications from our scientific visualization research. We further show that, despite these shared commonalities, there exists a variety of significantly different approaches to improve the responsiveness in the context of parallel and distributed hardware architectures. This is exemplified by means of several of our visualization research projects.
Stacks Image 957
Visualization of Temporal Similarity in Field Data
S. Frey, F. Sadlo and T. Ertl
Large parts of science and engineering deal with time-dependent phenomena. We present an interactive visualization approach for detecting and exploring similarity in the temporal variation of field data. It allows the investigation of periodic and quasi-periodic behavior at single points as well as similarity between different locations within a field or between different data sets.
Stacks Image 1155
SIMT Microscheduling: Reducing Thread Stalling in Divergent Iterative Algorithms
S. Frey, G. Reina and T. Ertl
Current GPUs execute group of threads (warps) in lockstep. This potential leads to a large amount of wasted cycles for divergent control flow. To overcome this issue, we propose techniques to relax divergence on the fly within a computation kernel to achieve a much higher utilization of processing cores.
Stacks Image 1238
GPU-Accelerated Visualization
Ament, Marco; Frey, Steffen; Müller, Christoph; Grottel, Sebastian; Ertl, Thomas; Weiskopf, Daniel
In Book: E. W. Bethel, H. Childs, and C. Hansen: High Performance Visualization: Enabling Extreme-Scale Scientific Insight. Chapman and Hall/CRC (2012).
The book explores several distinct but interrelated approaches to high performance visualization. In my section, I give an overview on programming frameworks for GPU clusters with focus on visualization.
Stacks Image 739
Loose Capacity-Constrained Representatives for the Qualitative Visual Analysis in Molecular Dynamics
S. Frey, T. Schlömer, S. Grottel, C. Dachsbacher, O. Deussen and T. Ertl
The increasing extents of the spatial and temporal domain of molecular dynamics simulations pose a particular challenge for the visualization. Our technique replaces the huge amount of simulated particles by a smaller set of representatives that capture the characteristics of the underlying particle density and exhibit coherency over time.
Stacks Image 929
Load Balancing Utilizing Data Redundancy in Distributed Volume Rendering
S. Frey and T. Ertl
In distributed volume rendering, the cost for rendering different blocks of the volumes strongly varies with the camera configuration. Traditional load-balancing induces expensive data transfers. Our technique stores volume blocks redundantly, allowing our scheduler to evenly balance the load with almost no overhead.
Stacks Image 445
GPU-based 2D Flow Simulation Steering using Coherent Structures
M. Ament, S. Frey, F. Sadlo, T. Ertl and D. Weiskopf
The interactive investigation of CFD flow can both allow to achieve a desired flow behavior faster and supports the understanding of underlying mechanisms. We propose a CUDA-based steering system that allows interactive manipulation of boundary conditions such as obstacles or velocity profiles.
Stacks Image 1404
DIANA: A Device Abstraction Framework for Parallel Computations
A. Panagiotidis, D. Kauker, S. Frey, T. Ertl
There is a multitude of different APIs, SDKs and libraries for programming different many-core devices. DIANA provides a common interface to hide the complexity of managing them, allowing for easier maintainability, higher flexibility and improved portability.
Stacks Image 697
Interactive High-Quality Visualization of Higher-Order Finite Elements
M. Üffinger, S. Frey, and T. Ertl
Eurographics 2010 (EG'10), Norköpping, Sweden, 2010.
Higher-order finite element methods produce complex grids which feature non-convex, curvilinear cells with varying polynomial degree. We introduce a distributed GPU-based ray casting system employing both adaptive sampling and load-balancing for achieving interactive frame rates.
Stacks Image 655
Memory Saving Fourier Transform on GPUs
D. Kauker, H. Sanftmann, S. Frey, and T. Ertl
Current GPU Fourier transform libraries need a large buffer for storing intermediate results, severely limiting the size of an image that can be processed for instance. Our alternative two-dimensional Discrete Fourier Transform method computes the same output with far less memory by exploiting the separability property.
Stacks Image 1321
PaTraCo: A Framework Enabling the Transparent and Efficient Programming of Heterogeneous Compute Networks
S. Frey and T. Ertl
In particular ad-hoc compute networks are typically heterogeneous, e.g. different classes of compute devices at different speeds suited for different kinds of tasks, varying network bandwidth). We propose a framework with a built-in scheduler that explicitly considers these characteristics and handles them transparently for the user.
Stacks Image 529
Accelerating Raycasting Utilizing Volume Segmentation of Industrial CT Data
S. Frey and T. Ertl
Raycasting large CT volumes interactively is still computationally demanding. Utilizing the segmentation information that is typically employed for object analysis, GPU ray casting can be accelerated significantly using a novel data structure that is integrated into the volume, requiring no extra texture lookups.
Stacks Image 927
Concurrent CT Reconstruction and Visual Analysis Using Hybrid Multi-resolution Raycasting in a Cluster Environment
S. Frey, C. Müller, M. Strengert, and T. Ertl
Preparing the data for analysis after the CT scanning of an object takes a considerable amount of time. Our distributed program architecture leverages all resources of a GPU cluster for the incremental reconstruction, segmentation and rendering, provide the user with continuously updated provisional results.
Stacks Image 781
CUDA-Accelerated Continuous 2-D Scatterplots
S. Bachthaler, S. Frey and D. Weiskopf
IEEE Visualization 2009 (VIS'09), Poster, Atlantic City, USA, 2009.
Continuous scatterplots represent a continous distribution function in a dense way in the scatterplot domain. We significantly speed up the original CPU approach with our GPU implementation to allow for interactive exploration.
Stacks Image 311
A Compute Unified System Architecture for Graphics Clusters Incorporating Data-Locality
C. Müller, S. Frey, M. Strengert, C. Dachsbacher and T. Ertl
CUDA is a parallel computing architecture for graphics cards. CUDASA logically extends its programming model and API for multi-GPU systems and distributed GPU computing. It includes an automatic GPU-accelerated scheduling mechanism that is aware of data locality to optimize GPU utilization.
  • Program Committee Member, ISC, 2017.
  • Chair, ISC Workshop on In Situ Visualization, 2016.
  • Program Committee Member, Symposium on Large Data Analysis and Visualization (LDAV), 2016.
  • Program Committee Member, SIGGRAPH ASIA 2016 Symposium on Visualization, 2016.
  • Program Committee Member, International Symposium on Visual Computing (ISVC), 2016.
Advised Student Bachelor/Master/Diploma Theses
  • Performance Quantification of Volume Visualization, Valentin Bruder, 2015/16.
  • Real-time Ray Tracing of Volumetric Data, Marcus Richter, 2015/16.
  • Dynamic Acceleration Structures for the Visualization of Time-Dependent Volume Data on the GPU, Hajun Jang, 2014/15.
  • Adaptive Frameless Raycasting for Interactive Volume Visualization, Constantin Weisser, 2014.
  • Extraction of High Quality Isosurfaces from Large Volume Data, Thomas Mezger, 2012/13.
  • Distributed Raytracing on GPU Clusters, Jochen Puff, 2010.
  • Algorithm Design and Algorithmic-Level Optimization of Video / Image Algorithm using an Abstract Common Interface for NVIDIA CUDA and Intel Larrabee Platforms, Daniel Kauker, 2009/2010.
  • Parallel Computation of Volumetric Illumination of Astrophysical Nebulae on GPU Clusters, Manuel Moser, 2009/2010.
  • Hauptseminar Advanced Visualization Techniques, 2012/13
  • Hauptseminar Volume Rendering, 2011/12
  • Seminar Interactive Visualization Techniques, 2011
  • Hauptseminar Volume Rendering, 2010
  • Lecture Assignments Visual Computing, 2009/10
  • Lecture Assignments Image Synthesis/Rendering, 2009
  • Lecture Assignments Visual Computing, 2008/09
  • Graphics and GPU programming lab, 2008/09
Stacks Image 843
GPU-basierte Kegelstrahlrekonstruktion großer CT Datensa╠łtze
Advisiors: Magnus Strengert (VISUS), Rolf Schaller (Daimler AG)
CT reconstruction plays an important role in industrial material testing and quality control. However, this process step can take considerable time, thus delaying the overall workflow.
The work introduces and evaluates several approaches for accelerating the Feldkamp CT reconstruction algorithm using a GPU.
Stacks Image 881
Depth Peeling in OpenSceneGraph
Depth peeling extracts 2D layers of 3D geometry in depth-sorted order. I implemented this technique as final project for the Graphics and GPU programming lab. I also added some sketchy drawing outline modes to generate the look of hand-drawn sketches. It was submitted to the OpenSceneGraph project and is contained in the package as the osgdepthpeeeling example.
Stacks Image 36
Webservice-Based Remote Rendering
I was working on a web-based service for rendering medical volumetric data sets as a research assistant of Friedemann Rößler at the VIS .
Stacks Image 915
Extending a Multimedia Engine for 3d madness GmbH
3d madness is a multimedia studio, with focus on the 3D Visualization of products, buildings and data as well ass virtual reality amongst others. I was working on making a powerful open source multimedia engine (Delta3D) accessible for the 3D modelers and designers as well as integrating effects (e.g. shadows) according to their needs.