Abstract
Graphical abstracts (GAs) are visual summaries that convey the key ideas, methods, and findings of academic papers at a glance. However, existing GA generation methods typically produce raster graphics that are difficult to post-edit and risk hallucinating or fabricating scientific data through image generation. We propose a framework for generating data-grounded GAs directly as editable vector graphics, enabling element-level editing in common drawing tools. We also introduce the Structural Independence Coefficient (SIC) to quantify editing simplicity. Experiments and a user study show that our method improves editability while preserving visual quality, accelerating reliable scientific communication within AI for Science.
Shunsuke Kitada, Ph.D.
Research Scientist working on Vision & Language with Deep Learning
My research interests include deep learning-based natural language processing, computer vision, medical image processing, and computational advertising.