arXiv subjects

Tianwei Ma

Publications and source records attributed to Tianwei Ma.

3 recordsLinked to original sources

arXiv2025

Characterizing Creativity in Data Visualization: Reflections and Future Directions

Characterizing creativity in visualization design can lead to the design of more expressive representations and visualization authoring tools that prioritize human creativity. In this paper, we examine how creativity manifests itself in visualization design processes through two complementary studies. First, a systematic review of 63 papers yields a design space spanning three themes: creative design frameworks that focus on developing design processes by incorporating divergent and convergent thinking activities, creative visual representations that focus on developing unorthodox visualizations, and visualization-enabled creativity support tools that focus on supporting a creative task (e.g., writing) with visualization. Second, we conducted qualitative interviews with 11 visualization practitioners and researchers to understand practical challenges and contrast those with current academic framing through our design space. The interview findings indicate that artifacts or final products (unorthodox visualizations) are often disproportionately considered as the primary indicator of creativity, whereas the design process remains undervalued in practical and organizational contexts. We also found that ideation is a universal bottleneck, and organizational constraints are often the primary barrier to creative work. We discuss implications for rethinking the relationship between our design space categories, addressing organizational barriers, and designing future frameworks, tools, and evaluation methods that better support creativity in the age of AI-assisted visualization. The full list of coded papers is available here: https://vizcreativity.notion.site/coded-papers.

cs.HC↗

arXiv2024

StoryNavi: On-Demand Narrative-Driven Reconstruction of Video Play With Generative AI

Manually navigating lengthy videos to seek information or answer questions can be a tedious and time-consuming task for users. We introduce StoryNavi, a novel system powered by VLLMs for generating customised video play experiences by retrieving materials from original videos. It directly answers users' query by constructing non-linear sequence with identified relevant clips to form a cohesive narrative. StoryNavi offers two modes of playback of the constructed video plays: 1) video-centric, which plays original audio and skips irrelevant segments, and 2) narrative-centric, narration guides the experience, and the original audio is muted. Our technical evaluation showed adequate retrieval performance compared to human retrieval. Our user evaluation shows that maintaining narrative coherence significantly enhances user engagement when viewing disjointed video segments. However, factors like video genre, content, and the query itself may lead to varying user preferences for the playback mode.

cs.HC↗

arXiv2023

SSIL: Self-Supervised Imitation Learning for End-to-End Driving

In autonomous driving, the end-to-end (E2E) driving approach that predicts vehicle control signals directly from sensor data is rapidly gaining attention. To learn a safe E2E driving system, one needs an extensive amount of driving data and human intervention. Vehicle control data is constructed by many hours of human driving, and it is challenging to construct large vehicle control datasets. Often, publicly available driving datasets are collected with limited driving scenes, and collecting vehicle control data is only available by vehicle manufacturers. To address these challenges, this paper proposes the first self-supervised learning framework, Self-Supervised Imitation Learning (SSIL), for E2E driving. The proposed SSIL framework can learn vision-based E2E driving networks without using driving command data or a pre-trained model. To construct pseudo steering angle data, proposed SSIL predicts a pseudo target from the vehicle's poses at the current and previous time points that are estimated with light detection and ranging sensors. In addition, we propose a new cross-attention-based conditioning approach (CACA) for a vision encoder in E2E driving, where a high-level instruction serves as the conditioning signal for visual information. Our numerical experiments with three different benchmark datasets demonstrate that the proposed SSIL framework achieves very comparable E2E driving accuracy with the supervised learning counterpart. Furthermore, the proposed pseudo-label predictor outperformed an existing one using proportional integral derivative controller, and proposed CACA achieved superior performance over existing conditioning approaches.

cs.RO↗