716
Aggregation and Multiscale Techniques
Principal Investigator | Scientific Staff | Technical Staff | |
|---|---|---|---|
Nobelstr. 15 70569 Stuttgart 0711/7816-368 | Universitätsstr. 38 70569 Stuttgart 0711/7816-363 | Universitätsstr. 38 70569 Stuttgart 0711/7816-264 | |
The long-term goal of this sub project is the development of efficient methods for computation, representation, processing and visualization of aggregated particle-oriented data on multiple scales. Hybrid particle-/continuum simulations and visual analysis of complex simulation data will be supported.
Short-term goals of the project include the development of hierarchical data structures for multiscale representations, aggregation of particle-based data to continuum-based data, the respective scalable visualization as well as interactive exploration of multi-attribute fields by utilizing multiple coordinated views. Special attention will be given to huge time-based data sets.
- Konzept der Multiskalenaggregation von Teilchendaten
(Kopie 1)
To close the gap between particle-based models typically used in SFB 716 and continuum-based field representations which are common to scientific visualization and grid-based simulations we employ aggregation methods.
Here, aggregation means a generalized summarized description of particles in a certain space region (e.g. the compution of pressure and temperature from the movement of particle groups). From a physics-bases point of view, aggregation allows the transition from the particle-based model of statistical physics to significant thermodynamic quantities that help to gain insight to the simulated phenomena.
The long-term research goals are the development of techniques for creating and processing multi-scale aggregation, representation of multi-scale and multi-attribute data as well as the interactive visualization of this data. For the multi-scale processing and visualization of huge data sets we have to take care of finding efficient methods. Similarly, the effectiveness of the visualization plays a big role and therefore we need user-oriented visual representations and interaction models - the multi-scale representation will offer different abstractions and a scalable visualization.
For the sub project, the following goals are set:
First, basic methods of aggregation of the original particle data are developed by employing physically-based aggregation models. Secondly, hierarchical data structures for multi-scale representation of the continuous fields are investigated. This representation should allow an efficient access to the data during the visualization process. Thirdly, visualization methods for multi-scale fields are surveyed. The focus lies on scalar, vectorial and tensorial fields as well as on simultaneous visualization of multiple scales to get different levels of abstraction of the data. Fourthly, interactive exploration of multi-attribute fields and original particle data with multiple coordinated views will be supported. It will be possible to perform brushing and linking in 2D scatter plots and 3D field visualizations as well as a drill-down to the original particle data. Fifthly, time-based data sets will be supported. Therefore, the resulting data that has to be processed increases significantly which requires suitable methods for interaction with the space-time data.
Essential strategies to handle the challenges of this sub project include hierarchical data structures for scalable algorithms, adaptive visualization techniques, out-of-core methods to process big data sets, multiple coordinated views and respective interaction methods as well as efficient GPU techniques (GPU: graphics processing unit).
Continuous Scatterplots
Conventional scatterplots are a well-established tool for visualizing multidimenionsal data sets. Data samples in such data sets usually have a spatial relationship which allows to interpolate between neigboring data samples. Visualizing these data sets with conventional scatterplots ignores the interpolated data, taking only discrete data samples into account.
We generalize the concept of scatterplots by visualizing continuously defined data in a continuous ("dense") scatterplot. The underlying mathematical model can be applied to data sets and scatterplots of arbitrary dimensionality. Usually, data sets are three-dimensional and are visualized in a 2-D scatterplot. We implemented continous scatterplots for data sets of 3-D tetrahedralized grids which allow linear interpolation for the computation of a 2-D continuous scatterplot.
Further sample screenshots can be found on the project homepage.
- Figure 1: Conventional scatterplot (left image) for comparison with the continuous version. These scatterplots visualize the blunt-fin data set, which is defined on a curvilinear grid (resolution: 40x40x32).
Efficient and Adaptive Rendering of 2-D Continuous Scatterplots
The approach to compute continuous scatterplots was limited to linearly interpolated data. We extend the concept of continuous scatterplots to allow for a broad class of interpolation methods within the spatial grid. The algorithm is scalable and adaptive since it allows us to balance computation time and scatterplot quality.
Publications
Conference Papers
2009
Efficient and Adaptive Rendering of 2-D Continuous Scatterplots.
In: Proceedings of EuroVis 2009, pp. 743-750. 2009
2008
Continuous Scatterplots.
In: Proceedings Visualization / Information Visualization 2008, pp. 1428-1435. 2008
