In recent years, curved computer monitors have become a viable option for consumers. However, traditional real-time graphics pipelines expect a flat display surface, and most real-time applications, such as games and simulations, do not consider the actual geometry of the monitor during rendering. As a result, the synthesized images appear distorted and unnatural when viewed on a curved display.
Distortion correction methods for correcting the lens distortion in cameras and head-mounted displays can also be utilized in real-time rendering software for curved monitors. However, the final distortion observed on the curved display depends on the user's viewpoint. With head-tracking, accurate distortion correction can be performed, and perspective-correct projections can be produced.
In this thesis, we analyze various methods for generating correct renderings based on the user's viewpoint and the geometry of the monitor. Our experiments confirm that image-based methods provide the best overall performance with acceptable image quality. However, real-time ray tracing and geometry-based implementations are practicable alternatives when using current hardware, and these methods do not suffer from image resampling artifacts. Additionally, we present and evaluate a custom subdivision scheme as an alternative to hardware tessellation for geometry-based solutions that can be implemented in a single render pass using the recently introduced graphics mesh pipeline. In our subdivision scheme, the geometry is split along a screen-aligned grid that reflects the geometry of the display more accurately than the fixed tessellation patterns of hardware tessellation. While the performance of our software-based subdivision scheme has to be improved further, it produces fewer triangles for coarse geometry and, at the same time, achieves similar image quality to hardware tessellation.
Technological advances have dramatically expanded our ability to collect data of neural connectivity in the brain and apply this data in the field of connectomics. The focus of research is thus increasingly shifting towards the analysis of this complex data. Many applications visualize neurological data in three-dimensional space. However, these require interactivity to view hidden data and are not always applicable. To support neuroscientific research we present Spatial-Data-Driven Layouts, a novel web-tool to visualize neuronal networks of multiple species in two-dimensional space. Our method is data-driven and is therefore independent of species or perspective. We generate node-link diagrams where nodes represent brain regions, while the edges correspond the connectivity. To realize this data-driven approach we apply Parcellation-derived Connectivity, generated from We provide further guidance by visually encoding anatomical context of the underlying brain hierarchy. Colored parcellations in the background encapsulate and cluster nodes that belong to the same super-regions. Additionally the background provides an overall shape, similar to the brain and is independent of the graph’s completeness, facilitating the comparison of sub-networks with each other as well as with the entire network. The background is customizable in terms of anatomical details to reflect either the anatomical size or the number of connections per region.
We conduct case studies for two species, mouse and human, to validate our visualizations and show that the spatial distribution of nodes reflects the anatomy of the brain. Nodes are adjacent to each other if they also represent neighboring regions in the reference space.
The results provided by Spatial-Data-Driven Layouts are evaluated in a web-based user study involving domain experts in neuroscience, computer science, computational science, bioinformatics, and computational biology. Evaluating the studies for two different species, mouse and human, shows that our methodology can be applied data-driven and species-independent. The feedback obtained from the experts indicates clear potential. Spatial-Data-Driven Layouts quickly and easily recreate illustrations in literature that usually are created with a great deal of effort. Added context in sub-networks to preserve the overall shape of the brain and to make those networks comparable to each other, was considered very useful. Spatial-Data-Driven Layouts is a novelty in the visualization of neuronal circuits of the Drosophila melanogaster larval brain and considered a first good step in this direction.
In the future, we plan to extend the application with interactivity to provide neuroscientists with an intuitive representation of their data. The customization of brain regions, connectivity, as well as details of the layout via parameters, can be adapted to their interests. In addition, we aim to improve neuron-level visualization and visual encoding of the Drosophila larval network graphs to provide a more detailed representation of circuits.
20 + 10
Supervisor: Eduard Gröller
Institute of Visual Computing & Human-Centered Technology
Favoritenstr. 9-11 / E193-02
Austria - Europe