One Figure, Every Canvas: Editable Flowchart Relayout via Agentic Pipeline

Problem

This work addresses the challenge of aspect-ratio-adaptive flowchart relayout across various canvas formats. The existing literature lacks effective methods for maintaining content fidelity when adapting flowcharts to different aspect ratios. This paper is a preprint and has not undergone peer review.

Method

The authors propose an agentic pipeline consisting of three stages: Parse, Style, and Layout. The core algorithm features a main agent that collaborates with a critic to ensure connectivity checks throughout the relayout process. The output is formatted as editable mxGraph XML, allowing for easy manipulation of the flowcharts. The data used for training consists of a curated benchmark of 100 flowcharts, each evaluated at five different aspect ratios. Specific details regarding the loss function and training compute resources are not disclosed.

Results

The proposed method achieves a content fidelity score of 68.6%, significantly outperforming prior work, which reports fidelity scores ranging from 11.2% to 41.4%. This demonstrates a substantial improvement in the ability to maintain the integrity of flowchart content during aspect ratio adjustments.

Limitations

The authors do not report any limitations in their work. However, the absence of a specified loss function and training compute details may hinder reproducibility and further optimization of the proposed method.

Why it matters

The implications of this research are significant for downstream applications in automated diagram generation and editing tools. By enhancing the adaptability of flowcharts to various formats while preserving content fidelity, this work paves the way for more versatile and user-friendly diagramming solutions.