polyaug 0.1.4

Polygon topology augmentation CLI for segmentation datasets.

PolyAug

PolyAug is a Python package and CLI for topology-aware polygon augmentation for segmentation datasets.

Generating augmented datasets is common with existing libraries such as Albumentations and Augmentor; however, they do not reliably preserve polygon structure when the polygon forms a ring or path-like shape. Such structures are common in raster-to-vector tasks and datasets such as floorplans.

See the paper for more details: Topology-Preserving Data Augmentation for Ring-Type Polygon Annotations.

This package addresses this gap by mitigating the topology inconsistency in such processes.

For installation, usage, and development instructions, see the sections below.

---

Features

• Augments images and LabelMe JSON together
• Choose augmentation parameters and ranges, or use defaults
• Supports cropping, rotation, scaling, translation, brightness, contrast
• CLI and Python API for flexibility
• Generates LabelMe JSON annotations in standard and indexed formats

---

Installation

We recommend installing inside a virtual environment to avoid dependency conflicts.

Supports windows, macOS, and Linux.

Install from PyPI

python3 -m venv .venv
source .venv/bin/activate
pip install --upgrade pip

pip install polyaug

Quick CLI Check

After installation, verify the CLI is available:

polyaug --help

---

Basic CLI Usage

polyaug \
  --img-dir ./dataset/images \
  --json-dir ./dataset/json \
  --save-dir ./output \
  --num-per-image 2 \
  --crop 0.8 0.9 \
  --rotation -30 30 \
  --scale 0.7 1.3 \
  --translate -0.1 0.1 \
  --brightness -0.1 0.1 \
  --contrast -0.1 0.1

Inputs

• --img-dir: directory containing source images
• --json-dir: directory containing matching LabelMe JSON files

Outputs

Inside --save-dir, PolyAug writes:

• images/ → augmented images
• json/ → standard LabelMe JSON annotations
• augmented_index_json/ → indexed/debug JSON annotations

---

CLI Arguments

Required

• --img-dir
• --json-dir

Optional

• --save-dir
• --index-json-dir
• --num-per-image
• --crop MIN MAX
• --rotation MIN MAX
• --scale MIN MAX
• --translate MIN MAX
• --brightness MIN MAX
• --contrast MIN MAX
• --p-rotate
• --p-flip-h
• --p-flip-v
• --p-affine
• --p-crop
• --p-brightness
• --contour-simplify-epsilon
• --min-component-area
• --min-mask-pixel-area
• --random-aug-per-image
• --debug

To see the latest CLI options:

polyaug --help

---

Project Structure for Building the Package

A clean structure for the package is:

polyaug/
├── LICENSE
├── README.md
├── pyproject.toml
├── src/
│   └── polyaug/
│       ├── __init__.py
│       ├── augmentor.py
│       ├── cli.py
│       └── helper.py
└── tests/
    └── test_imports.py

---

Citation / Research Use

You can cite PolyAug as:

@software{laudari2026polyaug,
  title={PolyAug: Topology-Aware Polygon Augmentation for Segmentation Datasets},
  author={Laudari, Sudip},
  year={2026},
  version={0.1.4},
  url={https://pypi.org/project/polyaug/}
}

---

License

MIT License. See LICENSE for details.