EdgeFlow: Edge-Map Augmented VLM-Based Flowchart Processing for Industrial Requirements Engineering
Zhifei Dou, Shabnam Hassani, Ou Wei
Key claim
EdgeFlow significantly enhances flowchart-to-Mermaid conversion.
EdgeFlow improves the conversion of flowcharts to machine-readable models by using a Canny edge map as a structural prior. It achieves notable increases in node-level and edge-level F1 scores, demonstrating its effectiveness in industrial requirements engineering. This method does not require annotated training data, making it practical for real-world applications.
The proposed EdgeFlow method significantly enhances flowchart conversion using a novel structural prior.
The evaluation on a real-world dataset with substantial improvements supports the claims made.
Deep reliability assessment
The methodology supports the claim that EdgeFlow improves flowchart-to-Mermaid conversion by using Canny edge maps as structural priors, but the generalization to other diagram types or broader datasets is not fully validated.
Reproducibility
No open source code or dataset is provided, limiting reproducibility.
Discussion questions
- How does the reliance on Canny edge detection affect the generalizability of EdgeFlow to other types of diagrams?
- What are the practical implications of using EdgeFlow in environments with different levels of document noise?
- What specific conditions or datasets would demonstrate the limitations of EdgeFlow's approach?
Key figure
Figure 1 illustrates an example flowchart and its corresponding Canny edge map, highlighting the structural skeleton that aids in identifying cyclic topology.