Knossos and Ariadne: Benchmarking and Learning Complete Diagram Topology Extraction
What it is
This paper introduces Knossos, a large-scale benchmark of 19,200 diagrams with complete topology annotations, and Ariadne, a framework for extracting diagram structures. Ariadne decomposes the extraction task into node inventory and source-conditioned edge prediction to improve accuracy on complex diagrams.
The problem it solves
Existing vision-language models can recognize individual elements in diagrams but struggle to extract the complete topology, including all nodes and the specific relations/connectors between them.
What you could use it for
- Training vision-language models to convert engineering schematics into graph databases
- Automatically extracting logical flow from software architecture diagrams
- Parsing complex scientific or procedural diagrams into machine-readable formats
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.