The urgent need to improve health communication is highlighted by the millions of premature deaths worldwide each year due to lifestyle choices and behavioral risks. These losses reveal that researching and understanding these risks is not sufficient; we must also communicate them effectively to the public. In this paper, we discuss how we can assist experts in creating data-based risk visualizations for the general public. Our tool, RACCOON, is able to identify and suggest the most important risk factors in a data set, visualizing them in a way that allows seamless exploration of the data set. Then, we use the latest research in risk communication, narrative visualization, and affective visualization to generate engaging visualizations for the general public. Extensive customization options allow the expert to integrate their domain knowledge, and tailor the visualizations to their data story and communicative intent. We evaluated RACCOON with domain experts, as well as our visualizations with the general public. The findings highlight RACCOON's effectiveness in providing intuitive and engaging visualizations that appeal to a broad audience. They also provide first insights into the interplay of visualization design and communicative intent. By fusing the research fields of risk communication, narrative visualization, and affective visualization in one visualization generation tool, we provide a novel approach to support domain experts in communicating risks and risk factors to the general public.
Recirculation in flows is an important phenomenon of dynamical systems as it is linked to numerous further properties and behaviors. A formal definition of recirculation surfaces has been introduced in previous work. However, the extraction and visualization of such surfaces is a highly complex challenge as they are 2-manifolds in the 5D space. Although an approach for the geometry extraction exists, there are still several unsolved problems, mainly connected to the computational effort and surface reconstruction. In this work, we propose a fundamentally different idea: Instead of extracting an explicit geometry, we apply a direct ray tracing approach. This way, we effectively circumvent the challenge of reconstructing the geometry. Additionally, we implement multiple strategies for an efficient computation. Due to this, we are able to provide a visualization of recirculation surfaces in a fraction of the computation time of existing approaches.
Data items arranged into groups form partitions, and across time or through variation of grouping criteria, those partitions may change. While alluvial diagrams, showing the flow of data items as streams, visually capture such changes in partition sequences, their focus on showing similarities between neighboring partitions limits their application. Our paper introduces novel augmentations of alluvial diagrams with interactive visualizations and linked analysis, explicitly targeting the comparison of non-neighboring partitions without sacrificing the sequential nature of the data. Juxtaposed visualizations with the alluvial diagram's timeline provide a comparison of a selected partition to all other partitions, while additional scatterplot views provide an overview of the partition and set similarities. Connecting the set representations across views, we propose a coloring approach of sets and interactive selection mechanisms. The usefulness and generalizability of the approach are demonstrated through examples with application in supervised and unsupervised machine learning, as well as work collaboration analysis.
We present COMAND, a novel method for controllable multi-action 3D motion synthesis without requiring action-labeled data. Our method can generate a lifelike motion sequence containing consecutive non-locomotive actions such as kicking, jumping, or squatting, without the need for manual blending, enabling an intuitive way to control 3D human animation based on the desired motion types at specified time windows. At the core of our method is a motion manifold based on a periodic parameterization of a motion latent space that allows for unsupervised action clustering of 3D motion, thus allowing action-to-motion synthesis without the need to explicitly train the model on action-labeled datasets. This learned motion manifold has semantic and periodic properties that benefit 3D motion synthesis from action labels and from free-form text input, resulting in a state-of-the-art multi-modal and multi-action 3D motion generation framework. Our study shows that more than 83% and 96% of the users respectively rated COMAND as more natural and better matching the target action sequence when compared to existing methods.
Many visualization techniques exist for interactive exploration of segmentation volumes, however, photorealistic renderings are usually computed using slow offline techniques. We present a novel compression technique for segmentation volumes which enables interactive path tracing-based visualization for datasets up to hundreds of gigabytes: For every label, we create a grid of fixed-size axis aligned bounding boxes (AABBs) which covers the occupied voxels. For each AABB we first construct a sparse voxel octree (SVO) representing the contained voxels of the respective label, and then build a sparse voxel directed acyclic graph (SVDAG) identifying identical sub-trees across all SVOs; the lowest tree levels are stored as an occupancy bit-field. As a last step, we build a bounding volume hierarchy for the AABBs as a spatial indexing structure. Our representation solves a compression rate limitation of related SVDAG works as labels only need to be stored along with each AABB and not in the graph encoding of their shape. Our compression is GPU-friendly as hardware raytracing efficiently finds AABB intersections which we then traverse using a custom accelerated SVDAG traversal. Our method is able to path-trace a 113 GB volume on a consumer-grade GPU with 1 sample per pixel with up to 32 bounces at 108 FPS in a lossless representation, or at up to 1017 FPS when using dynamic level of detail.
We present a novel method for real-time 3D hydraulic erosion simulation of large-scale terrain. Existing hydraulic erosion methods based on heightmaps and the virtual pipes method are extended to multi-layered heightmaps that can represent more complex 3D features. Our new method for horizontal erosion is able to create overhangs, arches and to some extent caves by allowing water to erode horizontally adjacent bedrock, eventually splitting a column into two new columns. Additionally, we developed an iterative method for bedrock support check that efficiently prevents floating terrain and unrealistic overhangs by propagating bedrock connectivity while incorporating the weight of each column. We implement our method in CUDA, using only features that are also available in standard compute shaders. On a RTX 3070, the resulting method runs at approximately 6ms and 23ms per simulation step for resolutions of 2048(2) and 4096(2), respectively.
Genealogists study familial and ancestral relations in the temporal and historical context of their life events, including a geospatial context that studies the spread of families' origins, familial historical migration events, and varying relations between families and certain places over time. Genealogical data often constitute large, multimodal graphs encoding familial ties, which genealogist usually want to analyse using graphical representations. Geospatial information can supplement these relations, reflecting dates and locations of significant life events. To account for this additional information, we present Geospatial Topographic Attribute Maps (GeoTAMs). GeoTAMs extend Topographic Attribute Maps (TAMs), integrating a structural and temporal view on an ancestral graph with means to depict geospatial information about families and individuals. We employ a multi-view approach to represent temporal and spatial information in a genealogy graph. Evaluation results from a user study show that GeoTAMs support complex queries over graphical connections, time, and space.
Our work extends common pixelization techniques, enabling novel geometric pop-art stylization. We employ dedicated feature analysis to autonomously extract facial features, ensuring the best recognizability of persons and facial expressions in portraits. Additionally, our method includes automated content-related detail level extraction for scenic image content. Based on these detail levels, a hierarchical structure sets the basis for non-uniform pixelization. A joint optimization routine computes a reduced color palette alongside the coarse superpixel segmentation. We propose an adapted modification to common superpixel methods to handle non-uniform sized cells, maintaining a comparable level of detail while allowing for a coarser, more pixelated look. Additionally, this intermediate result serves as the basis for our geometric abstraction by eventually clustering polygonal shapes based on the pixelization. Colors and shapes are derived from the source image to capture and reproduce the most essential details for recognizable characters and facial expressions. We document the theoretical details of our method, discuss and elaborate the possible extensions. Provided results of the intermediate pixelization are compared qualitatively to related methods. Compared to other stylization methods, our resulting geometric abstractions are generated automatically, preserving a high level of relevant details from the source image. Unlike simple filtering techniques or learning-based stylization methods, our approach allows for the incorporation of user input to highlight features. Furthermore, our method stays true to the original image and results in scale-independent vector graphics, rendering it a valuable tool for artists and graphic designers.
We introduce a novel versatile approach to enhance the quality of grid-based neural scene representations. Grid-based scene representations model a scene by storing density and color features at discrete 3D points, which offers faster training and rendering than purely implicit methods such as NeRF. However, they require high-resolution grids to achieve high-quality outputs, leading to a significant increase in memory usage. Common standard grids with uniform voxel sizes do not account for the varying complexity of different regions within a scene. This is particularly evident when a highly detailed region or object is present, while the rest of the scene is less significant or simply empty. To address this we introduce a novel approach based on frequency domain transformations for finding the key regions in the scene and then utilize a 2-level hierarchy of grids to allocate more resources to more detailed regions. We also created a more efficient version of this concept, that adapts to a compact grid representation, namely TensoRF, which also works for very few training samples.
Ensembles of simulations are generated to capture uncertainties in the simulation model and its initialization. When simulating 3D spatial phenomena, the value distributions may vary from region to region. Therefore, visualization methods need to adapt to different types and shapes of statistical distributions across regions. In the case of normal distribution, a region is well represented and visualized by the means and standard deviations. In the case of multi-modal distributions, the ensemble can be subdivided to investigate whether sub-ensembles exhibit uni-modal distributions in that region. We, therefore, propose an interactive visual analysis approach for region-based visualization within a hierarchy of sub-ensembles. The hierarchy of sub-ensembles is created using hierarchical clustering, while regions can be defined using parallel coordinates of statistical properties. The identified regions are rendered in a hierarchy of interactive volume renderers. We apply our approach to two real-world simulation ensembles to show its usability.
We introduce a neural level of detail pipeline for use in a GPU path tracer based on a sparse volumetric representation derived from neural radiance fields. We pre-compute lighting and occlusion to train a neural radiance field which faithfully captures appearance and shading via image-based optimization. By converting the resulting neural network into an efficiently rendered representation, we eliminate costly evaluations at runtime and keep performance competitive. When applying our representation to certain areas of the scene, we trade a slight bias from gradient-based optimization and lossy volumetric conversion for highly anti-aliased results at low sample counts. This enables virtually noise-free and temporally stable results at low computational cost and without any additional post-processing, such as denoising. We demonstrate the applicability of our method to both individual objects and a challenging outdoor scene composed of highly detailed foliage.
In flow visualization, the depiction of line geometry in three-dimensional domains is often accompanied by occlusions. If there is a notion of which geometry is important to see, then a careful adjustment of the transparency is possible to ensure that irrelevant geometry is not occluding the meaningful structures, which is an inherently view-dependent problem. Past work in this line of research focused on the view-dependent adjustment of the transparency only and left the color channel open for the encoding of additional information. For a given viewpoint, the colormap could be set by the user, but once the view changes, the visible geometry is different and the colormap might no longer be utilizing its full color range. Thus, in this paper, we readjust the color transfer function to the new view, such that the colors of the colormap are utilized uniformly in the final image. To this end, a visibility histogram of all scalar values is recalculated and equalized on the GPU each frame. Further, past approaches required a set of lines that is not too dense, since the opacity optimization would otherwise fade out all lines similarly. For this reason, we incorporate a hierarchical line clustering for which we experimentally study the influence of distance metrics, linkage options, and representative choices. We apply the method in a number of scientific data sets, including examples from atmospheric sciences, aerodynamics, and electromagnetism.
This paper introduces a novel glyph-based design for music representation that leverages deep latent features to improve userdirected search for music discovery. We propose a system that combines a pre-trained neural network model for high-level music feature extraction with dimensionality-reduction methods for effective visual mapping of the intrinsic characteristics that help distinguishing a song. We provide a search-by-icon user interface (UI) that integrates glyph based on the neural features in combination with other novel navigation methods to achieve intuitive search and exploration. A detailed user study validates our approach, demonstrating its efficacy in enabling swift song clustering, identification, and retrieval. Our findings reveal that our visual representation not only speeds up the music searching process but also fosters increased user interaction with digital music libraries, representing a valuable contribution to the domain of music exploration and retrieval.
Axis breaks are used in charts, for example, to reduce whitespace, to accommodate outliers, or to show data at different scales. Proposed in the 1980s, axis breaks have not gotten much attention since then in terms of what characterizes "good" breaks, how many of them to introduce, and where to best place them? To answer these questions, we propose a five-step framework that specifies (1) the number of breaks, (2) their position, (3) the scaling of the resulting subaxes, (4) the "niceness" of the breaks, and (5) the formatting of the breaks. To apply this framework, we introduce a new metric, called skew, to quantify how unevenly distributed points are along an axis. Skew is then used as a cost function to formulate the search for optimal axis breaks as a clustering problem, which we solve by applying a dynamic k-means algorithm. We apply our framework specifically to Parallel Coordinate Plots and compare our algorithmic solution to established methods like percentile breaks and Jenks natural breaks. An interactive testbed to try our framework as well as its source code are made freely available.
Coloring is a fundamental yet time-consuming task in the 2D animation production pipeline. Traditional methods typically rely on frame-by-frame user interaction, leading to high user time and production costs. In this paper, we introduce PropColor, an interactive yet simple tool to propagate colors between adjacent frames of a hand-drawn animation. Starting with an initial frame colored by the user, our method propagates colors to neighboring frames based on the Delaunay triangulations computed from the sketch contours and the color hints. In addition to propagating colouring between frames, our method also associates a confidence score with each of them. This enables to identify the frames where user intervention is needed the most, either to validate the result or to provide additional color hints. Experiments show that our lightweight tool gives real-time feedback and significantly cuts down the animator's time.
Interactive photorealistic rendering of 3D anatomy is used in medical education to explain the structure of the human body. It is currently restricted to frontal teaching scenarios, where even with a powerful GPU and high-speed access to a large storage device where the data set is hosted, interactive demonstrations can hardly be achieved. We present the use of novel view synthesis via compressed 3D Gaussian Splatting (3DGS) to overcome this restriction, and to even enable students to perform cinematic anatomy on lightweight and mobile devices. Our proposed pipeline first finds a set of camera poses that captures all potentially seen structures in the data. High-quality images are then generated with path tracing and converted into a compact 3DGS representation, consuming < 70 MB even for data sets of multiple GBs. This allows for real-time photorealistic novel view synthesis that recovers structures up to the voxel resolution and is almost indistinguishable from the path-traced images
We propose a codec specifically designed for meshlet compression, optimized for rapid data-parallel GPU decompression within a mesh shader. Our compression strategy orders triangles in optimal generalized triangle strips (GTSs), which we generate by formulating the creation as a mixed integer linear program (MILP). Our method achieves index buffer compression rates of 16:1 compared to the vertex pipeline and crack-free vertex attribute quantization based on user preference. The 15.5 million triangles of our teaser image decompress and render in 0.59 ms on an AMD Radeon RX 7900 XTX.
We present a new approach for video-driven animation of high-quality neural 3D head models, addressing the challenge of person-independent animation from video input. Typically, high-quality generative models are learned for specific individuals from multi-view video footage, resulting in person-specific latent representations that drive the generation process. In order to achieve person-independent animation from video input, we introduce an LSTM-based animation network capable of translating person-independent expression features into personalized animation parameters of person-specific 3D head models. Our approach combines the advantages of personalized head models (high quality and realism) with the convenience of video-driven animation employing multi-person facial performance capture. We demonstrate the effectiveness of our approach on synthesized animations with high quality based on different source videos as well as an ablation study.
A common way to handle boundaries in SPH fluid simulations is to sample the surface of the boundary geometry using particles. These boundary particles are assigned the same properties as the fluid particles and are considered in the pressure force computation to avoid a penetration of the boundary. However, the pressure solver requires a pressure value for each particle. These are typically not computed for the boundary particles due to the computational overhead. Therefore, several strategies have been investigated in previous works to obtain boundary pressure values. A popular, simple technique is pressure mirroring, which mirrors the values from the fluid particles. This method is efficient, but may cause visual artifacts. More complex approaches like pressure extrapolation aim to avoid these artifacts at the cost of computation time. We introduce a constraint-based derivation of Divergence-Free SPH (DFSPH) — a common state-of-the-art pressure solver. This derivation gives us new insights on how to integrate boundary particles in the pressure solve without the need of explicitly computing boundary pressure values. This yields a more elegant formulation of the pressure solver that avoids the aforementioned problems.
