DiagEx — reproducibility package for zero-shot P&ID-to-DEXPI digitization with a vision-language agent
Authors/Creators
- 1. ABB Corporate Research Center, Mannheim, Germany
Description
Reproducibility package for the manuscript DiagEx: Zero-Shot P&ID-to-DEXPI Digitization with a Vision-Language Agent (Koziolek, Braun, Bordis — under review). Contains the diagex Python package, the evaluation harness, ground-truth corpora for 10 public P&IDs, final results CSVs and figures for Phase 1 / Phase 2 / a no-tile ablation / a GPT-4.1 baseline, replay cassettes for offline re-derivation, and the manuscript PDF.
Per-run agent transcripts and intermediate DEXPI 2.0 JSON / SVG / tile crops are in a separate supplementary deposition on Zenodo, cross-linked via related_identifiers below — replace the placeholder DOI before publishing.
Files
hkoziolek/DiagEx-Repro-v1.0.2.zip
Additional details
Related works
- Is derived from
- https://github.com/hkoziolek/DiagEx-Repro (URL)
- Is supplement to
- Journal article: 10.0000/REPLACE_ME_PAPER_DOI (DOI)
- Is supplemented by
- Dataset: 10.5281/zenodo.19972466 (DOI)
Software
- Repository URL
- https://github.com/hkoziolek/DiagEx-Repro
References
- DEXPI Initiative (2025). DEXPI Specification 2.0.0. https://gitlab.com/dexpi/Specification (CC-BY 4.0).
- dexpi.specificator (DEXPI e.V., MIT). https://pypi.org/project/dexpi.specificator/