News & Resources
- Microsoft AI for Earth and SpaceNet Training Data Now Available on Radiant Earth Foundation’s Open Repository for Geospatial Training Data
- IEEE GRSS joins SpaceNet
- SpaceNet, Topcoder Crowdsourcing and Development Platform Partnership
Delivers Advanced Geospatial Data Science and Machine Learning Applications
- Capella Space Partners with SpaceNet® to Expand Access to SAR Data
- SpaceNet Roads Extraction and Routing Challenge Solutions are Released
- SpaceNet Team Earns the 2018 USGIF Industry Achievement Award
Recent Blog Posts
RESEARCH PAPERS & REPORTS
We publish research papers or reports that are made available through arXiv and any associated journal or official publication.
Our videos include presentations, event keynotes, podcasts, and more on our YouTube channel.
What is SpaceNet?
SpaceNet is a collaborative initiative that fosters research and innovation in the development of computer vision algorithms to automatically extract information from remote sensing data.
The four main goals of SpaceNet are to:
- Release openly licensed satellite imagery and associated geospatial labels
- Share open source software tools to help reduce the amount of geographic information system (GIS)expertise required of machine-learning developers interested in aerial and satellite imagery
- Publish and foster research on emerging analytical frameworks in open forums
- Sponsor open analytics prize challenges
What is the value of SpaceNet Challenges? How are the events useful?
- A significant amount of the world is not mapped. Organizations such as MissingMaps.org and Humanitarian OpenStreetMap Team help to solve this problem by providing tools and organizing mapping campaigns.
- The first step is to trace satellite imagery for building footprints and roads. From there, local knowledge can be applied to identify the buildings and roads.
- Prize competitions can help develop and assess technology that could be used to accelerate the mapping of these regions.
How are the SpaceNet Challenge results calculated?
The SpaceNet team works hard to establishes fair and geospatially meaningful metrics. When necessary, the team will create a new metric such as for the Road Network Extraction challenge. That challenge used the Average Path Length Similarity (APLS) Metric, created by Adam Van Etten, CosmiQ Works’ Senior Research Scientist. It is designed to assess network similarity with an emphasis on correct routing between nodes on the graph. For more information you can read Adam’s blog posts on Medium at The DownLinQ.
Each challenge has a detailed explanation of the metric used.
How has SpaceNet been used?
We have seen a large amount of use of the SpaceNet datasets, this has included published papers and internal company research and the support of multiple additional competitions
The winning algorithms have informed deployment of novel machine learning algorithms for both building footprint extraction and road network extraction.