Hi! Thanks for releasing AutoFigure. From the README, AutoFigure focuses on an iterative review–refine loop for generating editable SVG/XML figures from long-form text/papers.
We recently released DiagramBank (≈89k schematic diagrams + caption/abstract/in-text context) along with a ready-to-use RAG retrieval pipeline (indexing + top-k retrieval). We’re curious whether adding exemplar retrieval (retrieve similar methodology diagrams + their context as in-context references) could improve:
- structural fidelity (components/edges correctness),
- style/layout consistency,
- fewer refinement iterations / lower cost.
Would you be interested in a small ablation study or integration discussion? We can share a minimal adapter that returns top-k exemplars (images + caption/context) given a method section or caption.
Links:
DiagramBank repo: https://github.com/csml-rpi/DiagramBank
Preprint: https://www.researchsquare.com/article/rs-8917857/v1
Hi! Thanks for releasing AutoFigure. From the README, AutoFigure focuses on an iterative review–refine loop for generating editable SVG/XML figures from long-form text/papers.
We recently released DiagramBank (≈89k schematic diagrams + caption/abstract/in-text context) along with a ready-to-use RAG retrieval pipeline (indexing + top-k retrieval). We’re curious whether adding exemplar retrieval (retrieve similar methodology diagrams + their context as in-context references) could improve:
Would you be interested in a small ablation study or integration discussion? We can share a minimal adapter that returns top-k exemplars (images + caption/context) given a method section or caption.
Links:
DiagramBank repo: https://github.com/csml-rpi/DiagramBank
Preprint: https://www.researchsquare.com/article/rs-8917857/v1