# marine-pid-digitization

**Repository Path**: dlml2/marine-pid-digitization

## Basic Information

- **Project Name**: marine-pid-digitization
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-09-09
- **Last Updated**: 2026-09-09

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Marine P&ID Digitization Tool

Converts PDF/image-based Piping & Instrumentation Diagrams into a queryable graph database that answers:

- **"How do I line up from Tank A to Tank B?"**
- **"Main pump failed — what's the alternative route?"**

---

## Project Status

**Current Phase**: Phase 1 — Data Preparation

---

## Quick Start

```bash
# 1. Create virtual environment (Python 3.11 required)
pyenv local 3.11.0
python -m venv venv
source venv/bin/activate

# 2. Install dependencies
pip install -r requirements.txt

# 3. Export a P&ID page as a high-res JPG (300 DPI), place in data/raw/

# 4. Slice into tiles
python scripts/tile_generator.py --input data/raw/hfo_system.jpg --split

# 5. Label tiles in LabelImg
#    - Open Dir: data/tiles/images/train/
#    - Change Save Dir: data/tiles/labels/train/
#    - Format: YOLO
labelImg

# 6. Train YOLOv8
python scripts/train_yolo.py

# 7. Detect symbols + extract OCR tags
python scripts/ocr_extract.py --tiles data/tiles/images/train/ --page hfo_system

# 8. Import into Neo4j
python scripts/graph_import.py --all
```

---

## Folder Structure

```
Pipeline processing/
├── README.md
├── requirements.txt
│
├── config/
│   ├── classes.txt          ← 25 marine symbol classes
│   └── data.yaml            ← YOLOv8 dataset config
│
├── data/
│   ├── raw/                 ← Original high-res JPG exports from PDF
│   ├── tiles/
│   │   ├── images/
│   │   │   ├── train/       ← 80% of labeled tiles (LabelImg Open Dir)
│   │   │   └── val/         ← 20% validation tiles
│   │   └── labels/
│   │       ├── train/       ← YOLO .txt label files (LabelImg Save Dir)
│   │       └── val/
│   └── exports/
│       ├── tile_coordinate_log.csv   ← Maps tile → source page + offsets
│       └── component_maps/           ← Per-page JSON (output of ocr_extract.py)
│
├── models/
│   └── weights/
│       └── pid_detector_best.pt      ← Best trained model weights
│
├── scripts/
│   ├── tile_generator.py    ← Phase 1: Slice large P&IDs into tiles
│   ├── train_yolo.py        ← Phase 2: Fine-tune YOLOv8
│   ├── ocr_extract.py       ← Phase 3: Detect symbols + read tags
│   └── graph_import.py      ← Phase 4: Load into Neo4j
│
├── notebooks/               ← Jupyter notebooks for experimentation
├── graph/                   ← Neo4j query examples and Cypher scripts
├── outputs/                 ← Final reports, line-up procedure exports
└── plans/
    ├── rough.md             ← Original Gemini conversation notes
    └── PID_Digitization_Plan.md  ← Full structured implementation plan
```

---

## Neo4j Queries

After importing, open Neo4j Browser at `http://localhost:7474` and run:

```cypher
-- Find all pumps
MATCH (c:Component {type: "Pump_Centrifugal"}) RETURN c

-- Shortest path between two tanks (line-up procedure)
MATCH p = shortestPath(
  (a:Component {id: "TANK_HFO_BUNKER_1"})-[*]-(b:Component {id: "TANK_HFO_SETTLING"})
)
RETURN p

-- Alternative route when a pump fails
MATCH p = shortestPath(
  (a:Component)-[*]-(b:Component)
)
WHERE NONE(n IN nodes(p) WHERE n.id = "Pump_Centrifugal_0023")
RETURN p
```

---

## Key References

- [YOLOv8 Docs](https://docs.ultralytics.com)
- [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)
- [Neo4j Community Edition](https://neo4j.com/download/)
- [SAHI — Small Object Detection](https://github.com/obss/sahi)
- Full plan: `plans/PID_Digitization_Plan.md`