EuroVisShort2021

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/2633047

Browse

Now showing 1 - 20 of 21

VATUN: Visual Analytics for Testing and Understanding Convolutional Neural Networks
(The Eurographics Association, 2021) Park, Cheonbok; Yang, Soyoung; Na, Inyoup; Chung, Sunghyo; Shin, Sungbok; Kwon, Bum Chul; Park, Deokgun; Choo, Jaegul; Agus, Marco and Garth, Christoph and Kerren, Andreas
Convolutional neural networks (CNNs) are popularly used in a wide range of applications, such as computer vision, natural language processing, and human-computer interaction. However, testing and understanding a trained model is difficult and very time-consuming. This is because their inner mechanisms are often considered as a 'black-box' due to difficulty in understanding the causal relationships between processes and results. To help the testing and understanding of such models, we present a user-interactive visual analytics system, VATUN, to analyze a CNN-based image classification model. Users can accomplish the following four tasks in our integrated system: (1) detect data instances in which the model confuses classification, (2) compare outcomes of the model by manipulating the conditions of the image, (3) understand reasons for the prediction of the model by highlighting highly influential parts from the image, and (4) analyze the overall what-if scenarios when augmenting the instances for each class. Moreover, by combining multiple techniques, our system lets users analyze behavior of the model from various perspectives. We conduct a user study of an image classification scenario with three domain experts. Our study will contribute to reducing the time cost for testing and understanding the CNN-based models in several industrial areas.
EuroVis 2021 Short Papers: Frontmatter
(The Eurographics Association, 2021) Agus, Marco; Garth, Christoph; Kerren, Andreas; Agus, Marco and Garth, Christoph and Kerren, Andreas
A Multilevel Approach for Event-Based Dynamic Graph Drawing
(The Eurographics Association, 2021) Arleo, Alessio; Miksch, Silvia; Archambault, Daniel; Agus, Marco and Garth, Christoph and Kerren, Andreas
The timeslice is the predominant method for drawing and visualizing dynamic graphs. However, when nodes and edges have real coordinates along the time axis, it becomes difficult to organize them into discrete timeslices, without a loss of temporal information due to projection. Event-based dynamic graph drawing rejects the notion of a timeslice and allows each node and edge to have its own real-valued time coordinate. Nodes are represented as trajectories of adaptive complexity that are drawn directly in the three-dimensional space-time cube (2D + t). Existing work has demonstrated clear advantages for this approach, but these advantages come at a running time cost. In response to this scalability issue, we present MultiDynNoS, the first multilevel approach for event-based dynamic graph drawing. We consider three operators for coarsening and placement, inspired by Walshaw, GRIP, and FM3, which we couple with an event-based graph drawing algorithm. We evaluate our approach on a selection of real graphs, showing that it outperforms timeslice-based and existing event-based techniques.
RoomCanvas: A Visualization System for Spatiotemporal Temperature Data in Smart Homes
(The Eurographics Association, 2021) König, Bastian; Limberger, Daniel; Klimke, Jan; Hagedorn, Benjamin; Döllner, Jürgen; Agus, Marco and Garth, Christoph and Kerren, Andreas
Spatiotemporal measurements such as power consumption, temperature, humidity, movement, noise, brightness, etc., will become ubiquitously available in both old and modern homes to capture and analyze behavioral patterns. The data is fed into analytics platforms and tapped by services but is generally not readily available to consumers for exploration due in part to its inherent complexity and volume. We present an interactive visualization system that uses a simplified 3D representation of building interiors as a canvas for a unified sensor data display. The system's underlying visualization supports spatial as well as temporal accumulation of data, e.g., temperature and humidity values. It introduces a volumetric data interpolation approach which takes 3D room boundaries such as walls, doors, and windows into account. We showcase an interactive, web-based prototype that allows for the exploration of historical as well as real-time data of multiple temperature and humidity sensors. Finally, we sketch an integrated pipeline from sensor data acquisition to visualization, discuss the creation of semantic geometry and subsequent preprocessing, and provide insights into our real-time rendering implementation.
Selection of Optimal Salient Time Steps by Non-negative Tucker Tensor Decomposition
(The Eurographics Association, 2021) Pulido, Jesus; Patchett, John; Bhattarai, Manish; Alexandrov, Boian; Ahrens, James; Agus, Marco and Garth, Christoph and Kerren, Andreas
Choosing salient time steps from spatio-temporal data is useful for summarizing the sequence and developing visualizations for animations prior to committing time and resources to their production on an entire time series. Animations can be developed more quickly with visualization choices that work best for a small set of the important salient timesteps. Here we introduce a new unsupervised learning method for finding such salient timesteps. The volumetric data is represented by a 4-dimensional non-negative tensor, X(t; x; y; z).The presence of latent (not directly observable) structure in this tensor allows a unique representation and compression of the data. To extract the latent time-features we utilize non-negative Tucker tensor decomposition. We then map these time-features to their maximal values to identify the salient time steps. We demonstrate that this choice of time steps allows a good representation of the time series as a whole.
Evaluating Interactive Comparison Techniques in a Multiclass Density Map for Visual Crime Analytics
(The Eurographics Association, 2021) Svicarovic, Lukas; Parra, Denis; Lobo, María Jesús; Agus, Marco and Garth, Christoph and Kerren, Andreas
Techniques for presenting objects spatially via density maps have been thoroughly studied, but there is lack of research on how to display this information in the presence of several classes, i.e., multiclass density maps. Moreover, there is even less research on how to design an interactive visualization for comparison tasks on multiclass density maps. One application domain which requires this type of visualization for comparison tasks is crime analytics, and the lack of research in this area results in ineffective visual designs. To fill this gap, we study four types of techniques to compare multiclass density maps, using car theft data. The interactive techniques studied are swipe, translucent overlay, magic lens, and juxtaposition. The results of a user study (N=32) indicate that juxtaposition yields the worst performance to compare distributions, whereas swipe and magic lens perform the best in terms of time needed to complete the experiment. Our research provides empirical evidence on how to design interactive idioms for multiclass density spatial data, and it opens a line of research for other domains and visual tasks.
Visual Analysis of the Relation Between Stiffness Tensor and the Cauchy-Green Tensor
(The Eurographics Association, 2021) Blecha, Christian; Hergl, Chiara; Nagel, Thomas; Scheuermann, Gerik; Agus, Marco and Garth, Christoph and Kerren, Andreas
Stress and strain tensors, two well-known quantities in mechanical engineering, are linked through a fourth-order stiffness tensor, which is not considered by many visualizations due to its complexity. Considering an orthotropic material, the tensor naturally decomposes into nine known material properties.We used fiber surfaces to analyze a data set representing a biological tissue. A sphere is pushed into the material to confirm the mathematical link as well as the possibility to extract highly deformed regions even if only the stiffness tensor is available.
SailVis: Reconstruction and Multifaceted Visualization of Sail Shape
(The Eurographics Association, 2021) Mu, Danfeng; Pieras, Marcos; Broekens, Douwe; Marroquim, Ricardo; Agus, Marco and Garth, Christoph and Kerren, Andreas
While sailing, sailors rely on their eyes to inspect the sail shape and adjust the configurations to achieve an appropriate shape for a certain the weather condition. Mastering this so-called trimming process requires years of experience since the visual inspection of the sail shape suffers from inaccuracies and many times are difficult to communicate verbally. Therefore, this research proposes a visual analysis tool that presents an accurate sail shape representation and supports sailors in investigating the optimal sail shape for certain weather conditions. In order to achieve our goals, we reconstruct the 3D sail shape from point clouds acquired by photogrammetry methods. For incomplete acquisitions we deform a complete template sail to estimate the missing parts. We designed a visualization dashboard for sailors to explore the 3D structure, 2D profiles and characteristics of the time-varying sail shape as well as analyze their relation to boat speed. The usability of the visualization tool is tested through a qualitative evaluation with two sailing experts. The result shows that the reconstruction and deformation of sail shape are plausible. Furthermore, the visualization dashboard has the potential to enhance sailors' comprehension of sail shape and provide insights towards optimal trimming.
Algorithmic Improvements on Hilbert and Moore Treemaps for Visualization of Large Tree-structured Datasets
(The Eurographics Association, 2021) Scheibel, Willy; Weyand, Christopher; Bethge, Joseph; Döllner, Jürgen; Agus, Marco and Garth, Christoph and Kerren, Andreas
Hilbert and Moore treemaps are based on the same named space-filling curves to lay out tree-structured data for visualization. One main component of them is a partitioning subroutine, whose algorithmic complexity poses problems when scaling to industry-sized datasets. Further, the subroutine allows for different optimization criteria that result in different layout decisions. This paper proposes conceptual and algorithmic improvements to this partitioning subroutine. Two measures for the quality of partitioning are proposed, resulting in the min-max and min-variance optimization tasks. For both tasks, linear-time algorithms are presented that find an optimal solution. The implementation variants are evaluated with respect to layout metrics and run-time performance against a previously available greedy approach. The results show significantly improved run time and no deterioration in layout metrics, suggesting effective use of Hilbert and Moore treemaps for datasets with millions of nodes.
DanceMoves: A Visual Analytics Tool for Dance Movement Analysis
(The Eurographics Association, 2021) Arpatzoglou, Vasiliki; Kardara, Artemis; Diehl, Alexandra; Flueckiger, Barbara; Helmer, Sven; Pajarola, Renato; Agus, Marco and Garth, Christoph and Kerren, Andreas
Analyzing body movement as a means of expression is of interest in diverse areas, such as dance, sports, films, as well as anthropology or archaeology. In particular, in choreography, body movements are at the core of artistic expression. Dance moves are composed of spatial and temporal structures that are difficult to address without interactive visual data analysis tools. We present a visual analytics solution that allows the user to get an overview of, compare, and visually search dance move features in video archives. With the help of similarity measures, a user can compare dance moves and assess dance poses. We illustrate our approach through three use cases and an analysis of the performance of our similarity measures. The expert feedback and the experimental results show that 75% to 80% of dance moves can correctly be categorized. Domain experts recognize great potential in this standardized analysis. Comparative and motion analysis allows them to get detailed insights into temporal and spatial development of motion patterns and poses.
Visualization of Uncertain Multivariate Data via Feature Confidence Level-Sets
(The Eurographics Association, 2021) Sane, Sudhanshu; Athawale, Tushar M.; Johnson, Chris R.; Agus, Marco and Garth, Christoph and Kerren, Andreas
Recent advancements in multivariate data visualization have opened new research opportunities for the visualization community. In this paper, we propose an uncertain multivariate data visualization technique called feature confidence level-sets. Conceptually, feature level-sets refer to level-sets of multivariate data. Our proposed technique extends the existing idea of univariate confidence isosurfaces to multivariate feature level-sets. Feature confidence level-sets are computed by considering the trait for a specific feature, a confidence interval, and the distribution of data at each grid point in the domain. Using uncertain multivariate data sets, we demonstrate the utility of the technique to visualize regions with uncertainty in relation to the specific trait or feature, and the ability of the technique to provide secondary feature structure visualization based on uncertainty.
VisMiFlow: Visual Analytics to Support Citizen Migration Understanding Over Time and Space
(The Eurographics Association, 2021) Scheidl, Andreas; Leite, Roger A.; Miksch, Silvia; Agus, Marco and Garth, Christoph and Kerren, Andreas
Multivariate networks are complex data structures, which are ubiquitous in many application domains. Driven by a real-world problem, namely the movement behavior of citizens in Vienna, we designed and implemented a Visual Analytics (VA) approach to ease citizen behavior analyses over time and space. We used a dataset of citizens' movement behavior to, from, or within Vienna from 2007 to 2018, provided by Vienna's city. To tackle the complexity of time, space, and other moving people's attributes, we follow a data-user-tasks design approach to support urban developers. We qualitatively evaluated our VA approach with five experts coming from the field of VA and one non-expert. The evaluation illustrated the importance of task-specific visualization and interaction techniques to support users' decision-making and insights. We elaborate on our findings and suggest potential future works to the field.
Discussion Flows: An Interactive Visualization for Analyzing Engagement in Multi-Party Meetings
(The Eurographics Association, 2021) Wang, Tao; Keck, Mandy; Vosough, Zana; Agus, Marco and Garth, Christoph and Kerren, Andreas
Engagement in multi-party meetings is a key indicator of outcome. Poor attendee involvement can hinder progress and hurt team cohesion. Thus, there is a strong motivation for organizations to better understand what happens in meetings and improve upon their experience. However, analyzing multi-party meetings is a challenging task, as one needs to consider both verbal exchanges and meeting dynamics among speakers. There is currently a lack of support on these unique tasks. In this paper, we present a new visual approach to help analyze multi-party meetings in industry settings: Discussion Flows, a multi-level interactive visualization tool. Its glyph-based overview allows effortless comparison of overall interactions among different meetings, whereas the individual meeting view uses flow diagrams to convey the relative participation of different speakers throughout the meeting agenda in different levels of details. We demonstrate our approach with meeting recordings from an open source dialogue corpora and use them as the benchmark dataset.
RISSAD: Rule-based Interactive Semi-Supervised Anomaly Detection
(The Eurographics Association, 2021) Deng, Jiahao; Brown, Eli T.; Agus, Marco and Garth, Christoph and Kerren, Andreas
Anomaly detection has gained increasing attention from researchers in recent times. Owing to a lack of reliable ground-truth labels, many current state-of-art techniques focus on unsupervised learning, which lacks a mechanism for user involvement. Further, these techniques do not provide interpretable results in a way that is understandable to the general public. To address this problem, we present RISSAD: an interactive technique that not only helps users to detect anomalies, but automatically characterizes those anomalies with descriptive rules. The technique employs a semi-supervised learning approach based on an algorithm that relies on a partially-labeled dataset. Addressing the need for feedback and interpretability, the tool enables users to label anomalies individually or in groups, using visual tools. We demonstrate the tool's effectiveness using quantitative experiments simulated on existing anomaly-detection datasets, and a usage scenario that illustrates a real-world application.
Toward an Interactive Voronoi Treemap for Manual Arrangement and Grouping
(The Eurographics Association, 2021) Abuthawabeh, Ala; Aupetit, Michael; Agus, Marco and Garth, Christoph and Kerren, Andreas
Interactive spatial arrangement and grouping (A&G) of images is a critical step of the sense-making process. We argue that to support A&G tasks, a visual encoding idiom should avoid clutter, show groups explicitly, and maximize the use of space while allowing free positioning. None of the existing interactive idioms supporting A&G tasks optimizes all these criteria at once. We propose and implement an interactive Voronoi treemap for A&G that fulfills all these requirements. The cells representing groups or objects can be dragged or clicked to arrange objects and groups and to create, merge, split, expand, or collapse groups. We present a usage scenario for an art quiz game and a comparative analysis of our approach to the recent Piling.js library for a categorization task of HiC data images. We discuss limitations and future work.
Integration-Aware Vector Field Super Resolution
(The Eurographics Association, 2021) Sahoo, Saroj; Berger, Matthew; Agus, Marco and Garth, Christoph and Kerren, Andreas
In this work we propose an integration-aware super-resolution approach for 3D vector fields. Recent work in flow field superresolution has achieved remarkable success using deep learning approaches. However, existing approaches fail to account for how vector fields are used in practice, once an upsampled vector field is obtained. Specifically, a cornerstone of flow visualization is the visual analysis of streamlines, or integral curves of the vector field. To this end, we study how to incorporate streamlines as part of super-resolution in a deep learning context, such that upsampled vector fields are optimized to produce streamlines that resemble the ground truth upon integration. We consider common factors of integration as part of our approach - seeding, streamline length - and how these factors impact the resulting upsampled vector field. To demonstrate the effectiveness of our approach, we evaluate our model both quantitatively and qualitatively on different flow field datasets and compare our method against state of the art techniques.
Analytic Ray Splitting for Controlled Precision DVR
(The Eurographics Association, 2021) Weiss, Sebastian; Westermann, Rüdiger; Agus, Marco and Garth, Christoph and Kerren, Andreas
For direct volume rendering of post-classified data, we propose an algorithm that analytically splits a ray through a cubical cell at the control points of a piecewise-polynomial transfer function. This splitting generates segments over which the variation of the optical properties is described by piecewise cubic functions. This allows using numerical quadrature rules with controlled precision to obtain an approximation with prescribed error bounds. The proposed splitting scheme can be used to find all piecewise linear or monotonic segments along a ray, and it can thus be used to improve the accuracy of direct volume rendering, scale-invariant volume rendering, and multi-isosurface rendering.
TaskVis: Task-oriented Visualization Recommendation
(The Eurographics Association, 2021) Shen, Leixian; Shen, Enya; Tai, Zhiwei; Song, Yiran; Wang, Jianmin; Agus, Marco and Garth, Christoph and Kerren, Andreas
General visualization recommendation systems typically make design decisions of the dataset automatically. However, these systems are only able to prune meaningless visualizations but fail to recommend targeted results. In this paper, we contributed TaskVis, a task-oriented visualization recommendation approach with detailed modeling of the user's analysis task. We first summarized a task base with 18 analysis tasks by a survey both in academia and industry. On this basis, we further maintained a rule base, which extends empirical wisdom with our targeted modeling of analysis tasks. Inspired by Draco, we enumerated candidate visualizations through answer set programming. After visualization generation, TaskVis supports four ranking schemes according to the complexity of charts, coverage of the user's interested columns and tasks. In two user studies, we found that TaskVis can well reflect the user's preferences and strike a great balance between automation and the user's intent.
Graceful Degradation for Real-time Visualization of Streaming Geospatial Data
(The Eurographics Association, 2021) Rafael, João; Moreira, João; Mendes, Daniel; Alves, Mário; Gonçalves, Daniel; Agus, Marco and Garth, Christoph and Kerren, Andreas
The availability of devices that can record locations and are connected to the Internet creates a huge amount of geospatial data that are continuously streamed. The informative visualization of such data is a challenging problem, given their sheer volume, and the real-time nature of the incoming stream. A simple approach like plotting all datapoints would generate visual noise, and not scale well. To tackle this problem, we have developed a visualization technique based on graceful degradation along three overlaid time periods (ongoing, recent, and history), each with a different visual idiom. A usability test of the proposed technique showed promising results.
Loss-contribution-based in situ Visualization for Neural Network Training
(The Eurographics Association, 2021) Lee, Teng-Yok; Agus, Marco and Garth, Christoph and Kerren, Andreas
This paper presents an in situ visualization algorithm for neural network training. As each training data item leads to multiple hidden variables when being forward-propagated through a neural network, our algorithm first estimates how much each hidden variable contributes to the training loss. Based on linear approximation, we can approximate the contribution mainly based on the forward-propagated value and the backward-propagated derivative per hidden variable, both of which are available during the training with no cost. By aggregating the loss contribution of hidden variables per data item, we can detect difficult data items that contribute most to the loss, which can be ambiguous or even incorrectly labeled. For convolution neural networks (CNN) with images as inputs, we extend the estimation of loss contribution to measure how different image areas impact the loss, which can be visualized over time to see how a CNN evolves to handle ambiguous images.

Browse

Browsing EuroVisShort2021 by Issue Date

Results Per Page

Sort Options