Reproduce “Automatic Information Extraction from Piping and Instrumentation Diagrams” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rohit Rahul, Shubham Paliwal, Monika Sharma, Lovekesh Vig
TCS Research developed an end-to-end automated system for extracting and structuring information from scanned Piping and Instrumentation Diagrams (P&IDs) using a hybrid approach of deep learning and traditional computer vision. The system achieved 90.1% accuracy for pipeline code detection, 100% for inlet/outlet detection, and F1-scores over 0.86 for symbol segmentation on real-world P&ID sheets.
Key insights
- A hybrid approach effectively leverages the strengths of deep learning for complex visual patterns (text, symbols) and traditional computer vision for structured geometric elements (lines, polygons) in P&IDs.
- Fully Convolutional Networks (FCNs) can achieve high F1-scores (over 0.86) for semantic segmentation of P&ID symbols, even when faced with low inter-class visual distinctions and noisy data.
- Representing the schematic flow as a forest of trees, rooted at outlets and terminating at inlets, coupled with a pruning mechanism, provides an effective way to structure and validate complex connectivity information.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.