SciDiagramEdit: Learning to Edit Scientific Diagrams from Paper Revisions
SciDiagramEdit is a benchmark and skill evolution framework that mines before-and-after diagram pairs from arXiv paper version history and edits scientific diagrams based on natural language revision intents.
SciDiagramEdit is a benchmark and skill evolution framework that automatically edits scientific diagrams by learning natural language instructions from paper revisions. It extracts before-and-after diagram pairs from arXiv version history and uses agentic learning to continuously optimize editing skills through skill evolution.
The framework iteratively refines skill specifications from execution trajectories via an agentic proposer, potentially improving diagram editing accuracy. Next verifiable signal: improvement in editing accuracy on a held-out validation set.
This work demonstrates the potential of AI to automate tedious diagram editing in scientific workflows, potentially accelerating the paper revision process. Next verifiable signal: whether other teams adopt similar methods or integrate them into paper writing tools.
This technology can be embedded in paper writing or typesetting software, reducing researchers' manual diagram editing time and improving research efficiency. Next verifiable signal: whether a commercial company adopts the technology or launches a related product.
Future work may extend to more complex diagram types or integrate with other scientific automation tools. Next verifiable signal: whether open-source code or API is released.