Raster Feature Extractor

A CLI tool and Python library for extracting vector features from geospatial raster (TIF) files using Meta AI's Segment Anything Model (SAM), and exporting them as GeoJSON.

Installation

pip install orthomasker

Note: Installation using pip will fail in environments lacking a previous installation of the GDAL library, which is notoriously difficult to install using pip. Instead, using conda is generally recommended:

• conda create -n your_env_name python=3.10 gdal -c conda-forge
• conda activate your_env_name
• pip install orthomasker

Demo

Usage

# Using CLI
orthomasker your_input_filename.tif your_output_filename.geojson

# Using Python
from orthomasker.feature_extractor import RasterFeatureExtractor

# Provide your own test TIF file (upload or use a sample)
input_tif = "your_input_filename.tif"
output_geojson = "your_output_filename.geojson"

# Set up the extractor (use the path to your .pth file)
extractor = RasterFeatureExtractor()

extractor.convert(input_tif, output_geojson)

Options

• --sam-checkpoint: Path to SAM model weights (default: sam_vit_h_4b8939.pth)

• --model-type: SAM model type (vit_h, vit_l, vit_b; default: vit_h)

• --confidence-threshold: Minimum stability score to keep a mask (0–100; default: 0, no filter)

• --tile-size: Tile size for processing (default: 1024)

• --overlap: Tile overlap in pixels (default: 128)

• --class-name: Class label for output features (default: sam_object)

• --class-id: Class ID (e.g., 1) for output features (optional)

• --min-area: Minimum area (in square units of TIF CRS) for output features (optional)

• --max-area: Maximum area (in square units of TIF CRS) for output features (optional)

• --compactness: Minimum compactness threshold (0.0–1.0) using Polsby-Popper metric for filtering irregular shapes (optional)

• --fixed-bounds: Bounding box (minx, miny, maxx, maxy) in image CRS

• --merge: Merge overlapping polygons in output (optional)

• --verbose: Enable verbose output

Compactness Filtering

The --compactness option allows you to filter out irregular or elongated shapes by setting a minimum compactness threshold. This uses the Polsby-Popper compactness metric:

Compactness = (4π × Area) / (Perimeter²)

• Perfect circle: compactness = 1.0
• Square: compactness ≈ 0.785
• Elongated shapes: compactness approaches 0.0

• 0.1: Very permissive (removes only extremely irregular shapes)
• 0.3: Moderate filtering (removes highly irregular shapes)
• 0.6: Strict filtering (keeps only relatively compact shapes)
• 0.8: Very strict (keeps only very round/square shapes)

Acknowledgments

This project leverages Meta AI’s Segment Anything Model (SAM) for automatic mask generation; many thanks to Alexander Kirillov, et al. for their work!

Citations

@article{kirillov2023segany,
title={Segment Anything},
author={Kirillov, Alexander and Mintun, Eric and Ravi, Nikhila and Mao, Hanzi and Rolland, Chloe and Gustafson, Laura and Xiao, Tete and Whitehead, Spencer and Berg, Alexander C. and Lo, Wan-Yen and Doll{'a}r, Piotr and Girshick, Ross},
journal={arXiv:2304.02643},
year={2023}
}

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.