jpg-dwg 0.1.1

Convert scanned P&ID images to DXF/DWG

jpg-dwg

Convert scanned P&ID (Piping and Instrumentation) drawing images to DXF (lines only), and optionally to DWG via an external converter.

Install

From PyPI (recommended):

pip install jpg-dwg

From source:

git clone https://github.com/yourusername/jpg-dwg.git
cd jpg-dwg
pip install -e .

Or install dependencies only and run as module:

pip install -r requirements.txt
python -m jpg_dwg input.png -o output.dxf

Usage

# Basic: image → DXF (after pip install jpg-dwg)
jpg-dwg scan.png -o drawing.dxf

# Or as module
python -m jpg_dwg scan.png -o drawing.dxf

# With options (invert is on by default for black-on-white P&ID; use --no-invert if needed)
python -m jpg_dwg scan.png -o drawing.dxf --preview preprocessed.png --scale 0.1

# Enable P&ID symbol detection (contour-based heuristics)
python -m jpg_dwg scan.png -o drawing.dxf --symbols

# Also produce DWG (requires ODA File Converter)
python -m jpg_dwg scan.png -o drawing.dxf --dwg drawing.dwg

Options

Option	Description
input	Input image path (PNG, JPG, etc.)
-o, --output	Output DXF path
--scale	Drawing units per pixel (default 1.0)
--dpi	DPI hint (default 300)
--no-deskew	Disable automatic deskew
--invert	Invert binary (default: on for black lines on white)
--no-invert	Do not invert (for white lines on dark scan)
--adaptive	Use adaptive threshold instead of Otsu
--no-denoise	Disable morphology denoise
--thin	Thinning/skeletonization (requires opencv-contrib)
--preview PATH	Save preprocessed binary image
--symbols	Run P&ID symbol detection and insert blocks
--dwg PATH	Also convert DXF to DWG (see below)
-v, --verbose	Verbose logging

DWG output

This tool writes DXF only. To get DWG:

1. Install ODA File Converter.
2. Set ODA_FILE_CONVERTER to the converter executable path, or ensure it is on PATH.
3. Use --dwg output.dwg. The CLI will run the converter after writing the DXF.

Limitations

• Quality depends on scan resolution and contrast. Preprocess with --invert or --preview to tune.
• Vectorization extracts lines only (OpenCV Hough).
• P&ID symbol detection is heuristic (contour-based). For production, consider a trained model (e.g. YOLO on P&ID symbols).

Publishing to PyPI (maintainers)

To avoid distributing source code, build and upload only a bytecode-only wheel (no .py files, no sdist).

1. Create an account and API token at pypi.org (use test.pypi.org for testing).
2. From the repo root:

   pip install build twine
   python scripts/build_bytecode_wheel.py
   twine upload dist/*.whl
   

   • build_bytecode_wheel.py builds a wheel, then repacks it so the package contains only .pyc (no .py source). The resulting wheel in dist/ is suitable for upload.
   • Upload only the wheel (dist/*.whl); do not run python -m build and do not upload dist/*.tar.gz (sdist), so source is not published.
3. When prompted: username __token__, password = your PyPI API token.
4. To test first: twine upload --repository testpypi dist/*.whl, then pip install -i https://test.pypi.org/simple/ jpg-dwg.

Note: The bytecode-only wheel is built for the Python version that runs the script (e.g. 3.11). To support multiple versions, run scripts/build_bytecode_wheel.py under each desired Python (e.g. 3.9, 3.10, 3.11, 3.12) and upload all resulting wheels.

Update version in pyproject.toml and the [project.urls] Homepage/Source to your repo before publishing.

Project layout

jpg-dwg/
├── pyproject.toml
├── requirements.txt
├── README.md
├── scripts/
│   └── build_bytecode_wheel.py   # Build wheel with only .pyc (no source)
├── src/jpg_dwg/
│   ├── cli.py
│   ├── preprocess.py
│   ├── vectorize.py
│   ├── export_dxf.py
│   ├── symbols.py
│   └── config.py
└── tests/