Skip to main content

Datasets

Open Access

A dataset for detecting buildings, containers, and cranes in satellite images

Citation Author(s):
Hiroyuki Miyazaki
Submitted by:
Hiroyuki Miyazaki
Last updated:
DOI:
10.21227/7yfp-9p87
Average: 5 (1 vote)

Abstract

This dataset contains annotation image files representing shapes of buildings, containers, and cranes in satellite images on Google Earth. The datasets are structured in useful formats for machine learning and deep learning applications.

Instructions:

  • Buildings: The zip files contain PNG with annotations (a*.png; pixel value 1 for building and 0 for non-building) and JPEG files with satellite images (i*.jpg) in a size of 1024x1024 pixels with 0.298-m resolution. Three letters in file names indicate countries by ISO 3166 country codes.
  • Cranes: The zip files contain JPEG files of satellite images in 0.5-m resolution and LebelMe JSON with labeled polygons. The labels have classes of "crane", "gantry_crane", and "standby_gantry_crane."
  • Containers: The zip files contain JPEG files of satellite images in 0.5-m resolution and LebelMe JSON with labeled polygons.