GeoDataProcessor

Description

This library aims to help ML engineers to work with large satellite images. GeoDataProcessorv is fast python library that crops large satellite images files to small tiles for passing it to model. Library takes as an input directory with geo images and labels (as it shown in example of usage), makes clipping on input data and saves clipped tiles to saving_folder/images and saving_folder/labels respectively.

Example of usage

examples/preprocess_example.py

from pathlib import Path
from geodataset.fileutils.parse_directory import create_empty_folder
from geodataset.datasets import GeoImageDataset


def preprocess(images, labels, tile_size, saving_folder):
    dataset = GeoImageDataset(image_dataset=images,
                              shp_dataset=labels)

    dataset.clip_dataset(tile_size, output_directory=saving_folder)


def main():
    images = Path("data/images")
    labels = Path("data/labels")
    saving_folder = Path("buildings_train/")
    create_empty_folder(saving_folder)

    tile_size = 512
    preprocess(images, labels, tile_size, saving_folder)


if __name__ == '__main__':
    main()

Examples of input and output

Input image with labels	Cropped image with labels

Installing

locally:
- git clone https://github.com/homomorfism/GeoDataProcessor
- cd GeoDataProcessor
- pip install .
from pypl
- pip install GeoDataProcessor==1.0

Run tests

Run $ pytest

References

Data of segmentation of buildings were taken from aeronet tutorials

homomorfism/GeoDataProcessor