This repository accompanies the paper:
"Beyond Pixels: A Vector-to-Graph Framework for Reliable Schematic Auditing" (ICASSP 2025, under review)
Multimodal Large Language Models (MLLMs) exhibit strong pixel-level perception but suffer from structural blindness:
they fail to capture topology and symbolic logic in engineering schematics.
We propose Vector-to-Graph (V2G), a framework that:
- Parses CAD/DXF diagrams into property graphs (nodes = components, edges = connectivity).
- Uses an MLLM planner for compliance rule interpretation and subgraph selection.
- Employs Graph Signal Processing (GSP) verifiers for deterministic rule checking.
This pipeline makes structural dependencies explicit and enables reliable compliance auditing in power engineering.
cases/
Example schematic diagrams (DXF/PDF) and their corresponding JSON property graphs.examples/
Compliance checking results (e.g., grounding, wiring, labeling).docs/
Paper figures, benchmark description, and annotation guidelines.
👉 Code and database will be released after paper acceptance.
Currently, we provide case studies and JSON graph files for reproducibility of results.
- 60 real-world base cases collected from a regional power utility.
- Categories:
- Connection labeling errors
- Grounding errors
- Wiring errors
- Each case is augmented with 10–20 variants (rotation, translation, mild noise).
- Total size: ~900 instances.
⚡ This is the first diagnostic dataset for schematic auditing, released to encourage further study.
- Release sample cases (before acceptance).
- Release full benchmark (after acceptance).
- Release full implementation code (after acceptance).
- Release database schema for large-scale schematic auditing.
If you find this useful, please cite our paper:
@inproceedings{ma2025v2g,
title={Beyond Pixels: Vector-to-Graph Transformation for Reliable Schematic Auditing},
author={Ma, Chengwei and Zhou, Zhou and Xu, Zhixian and Zhu, Xiaowei and Hua, Xia and Shi, Si and Tian, Zhen and Yu, F. Richard},
booktitle={ICASSP},
year={2025}
}
