Published: Nov. 1, 2024
Language: Английский
Visual Informatics, Journal Year: 2024, Volume and Issue: 8(2), P. 43 - 66
Published: May 13, 2024
Generative AI (GenAI) has witnessed remarkable progress in recent years and demonstrated impressive performance various generation tasks different domains such as computer vision computational design. Many researchers have attempted to integrate GenAI into visualization framework, leveraging the superior generative capacity for operations. Concurrently, major breakthroughs like diffusion model large language also drastically increase potential of GenAI4VIS. From a technical perspective, this paper looks back on previous studies discusses challenges opportunities future research. Specifically, we cover applications types methods including sequence, tabular, spatial graph techniques which summarize four stages: data enhancement, visual mapping generation, stylization interaction. For each specific sub-task, illustrate typical concrete algorithms, aiming provide in-depth understanding state-of-the-art GenAI4VIS their limitations. Furthermore, based survey, discuss three aspects research evaluation, dataset, gap between end-to-end visualizations. By summarizing current limitations, endeavors useful insights
Language: Английский
Citations
23
Computer Science Review, Journal Year: 2025, Volume and Issue: 57, P. 100739 - 100739
Published: March 6, 2025
Language: Английский
Citations
0
Lecture notes in networks and systems, Journal Year: 2025, Volume and Issue: unknown, P. 89 - 97
Published: Jan. 1, 2025
Language: Английский
Citations
0
BIO Web of Conferences, Journal Year: 2025, Volume and Issue: 174, P. 03010 - 03010
Published: Jan. 1, 2025
MRG for medical images has become increasingly important due to the growing workload of radiologists in hospitals. However, current studies field predominantly focus on specific modal- ities or training foundation models with a notable lack research evaluating impact pre-trained performance across different tasks, particularly their cross-task capabilities. This study introduces novel benchmark multi-task learning that encompasses four modalities: CT, X-ray, ultrasound, and pathology. We believe this can provide robust comparative basis future field. More importantly, we conduct an in-depth analysis comparing modality-specific models, natural domain models. Our findings indicate generally outperform other all while exhibit superior cross-modality tasks. source code is available at https://github.com/Reckless0/MT-Med.git.
Language: Английский
Citations
0
International Journal of Geographical Information Science, Journal Year: 2024, Volume and Issue: unknown, P. 1 - 27
Published: Sept. 5, 2024
Language: Английский
Citations
3
IEEE Transactions on Visualization and Computer Graphics, Journal Year: 2024, Volume and Issue: 30(6), P. 3049 - 3061
Published: April 15, 2024
Extracting data insights and generating visual stories from tabular are critical parts of analysis. However, most existing studies primarily focus on stored as flat tables, typically without leveraging the relations between cells in headers hierarchical tables. When properly used, rich table can enable extraction many additional stories. To assist analysts storytelling, an approach is needed to organize these efficiently. In this work, we propose CoInsight, a system facilitate storytelling for tables by connecting insights. CoInsight extracts builds insight according structure headers. It further visualizes related using nested graph with edge bundling. We evaluate through usage scenario user experiment. The results demonstrate utility usability converting into
Language: Английский
Citations
2
IEEE Transactions on Visualization and Computer Graphics, Journal Year: 2024, Volume and Issue: 30(6), P. 2903 - 2915
Published: April 15, 2024
Temporal action localization aims to identify the boundaries and categories of actions in videos, such as scoring a goal football match. Single-frame supervision has emerged labor-efficient way train localizers it requires only one annotated frame per action. However, often suffers from poor performance due lack precise boundary annotations. To address this issue, we propose visual analysis method that aligns similar then propagates few user-provided annotations (e.g., boundaries, category labels) via generated alignments. Our models alignment between heaviest path problem annotation propagation quadratic optimization problem. As automatically alignments may not accurately match associated could produce inaccurate results, develop storyline visualization explain results their This facilitates users correcting wrong misalignments. The corrections are used improve other actions. effectiveness our improving is demonstrated through quantitative evaluation case study.
Language: Английский
Citations
2
IEEE Transactions on Visualization and Computer Graphics, Journal Year: 2024, Volume and Issue: 30(6), P. 3008 - 3021
Published: April 16, 2024
High-quality data is critical to deriving useful and reliable information. However, real-world often contains quality issues undermining the value of derived Most existing research on management focuses tabular data, leaving semi-structured under-exploited. Due schema-less hierarchical features discovering fixing challenging time-consuming. To address challenge, this paper presents JsonCurer, an interactive visualization system assist with in context JSON data. have overview issues, we first construct a taxonomy based interviews practitioners review 119 files. Then highlight schema that structural information, statistical features, Based similarity-based aggregation technique, depicts entire concise tree, where summary visualizations are given above each node, illustrated using Bubble Sets across nodes. We evaluate effectiveness usability JsonCurer two case studies. One domain analysis while other concerns assurance MongoDB documents. The source code available under Apache License 2.0 at https://github.com/changevis/JsonCurer.
Language: Английский
Citations
1
The International Journal of Advanced Manufacturing Technology, Journal Year: 2024, Volume and Issue: 135(5-6), P. 1893 - 1931
Published: Oct. 18, 2024
Language: Английский
Citations
1
IEEE Transactions on Visualization and Computer Graphics, Journal Year: 2023, Volume and Issue: 30(3), P. 1837 - 1852
Published: Dec. 21, 2023
Label quality issues, such as noisy labels and imbalanced class distributions, have negative effects on model performance. Automatic reweighting methods identify problematic samples with label issues by recognizing their validation assigning lower weights to them. However, these fail achieve satisfactory performance when the are of low quality. To tackle this, we develop Reweighter, a visual analysis tool for sample reweighting. The relationships between training modeled bipartite graph. Based this graph, improvement method is developed improve samples. Since automatic may not always be perfect, co-cluster-based graph visualization illustrate support interactive adjustments results. converted into constraints further We demonstrate effectiveness Reweighter in improving results through quantitative evaluation two case studies.
Language: Английский
Citations
2