PID_Symbol_Detection

Proposed Framework vs Conventional Framework

Benefits of Proposed Framework

Class-Agnostic Object Detection & One-shot Label Transfer is found to be more:

1. Generalizable to different underlying P&ID drawing styles
2. Robust to class-imbalance compared to equivalent class-aware counterparts.

Simplified Visual Walkthrough of Proposed Framework

1. Data preprocessing

This step breaks down large P&ID sheets into overlapping patches.

Plus, class-aware labels are transformed into class-agnostic to prepare for training a Yolo object detection model.

2. Train Yolo (Stage-1)

Trains a 'Generic' symbol detector

3. Inferencing with SAHI (Stage-1)

For large P&IDs infer on smaller patches and combine the results (implemented via SAHI ).

4. Label Transfer (Stage-2)

Train a model using one labeled image per symbol class (e.g. P&ID legend). The model can be a Siamese Network/ Prototypical (Zero-shot) Network or a Traditional classifier trained on augmented images.

If you use this package in your work, please cite it as:

@article{GUPTA2024105260,
title = {Semi-supervised symbol detection for piping and instrumentation drawings},
journal = {Automation in Construction},
volume = {159},
pages = {105260},
year = {2024},
issn = {0926-5805},
doi = {https://doi.org/10.1016/j.autcon.2023.105260},
url = {https://www.sciencedirect.com/science/article/pii/S0926580523005204},
author = {Mohit Gupta and Chialing Wei and Thomas Czerniawski},
}