Topcoder and SpaceNet: Solving Niche Challenges in Geospatial Mapping
What do you do when you want to leverage technology at scale to solve complex problems, but you have little to no support? It's a tough question. When most businesses and organizations get started in tech, they have a community. Typically, there's a network of other people around the globe trying to tackle the same (or similar) problems. You can connect, share data, and collaborate to build up new solutions iteratively.
But what if there isn't an existing networked community to help you? This is the problem that SpaceNet ran into head first when they started. At Topcoder Open 2019 (TCO19), SpaceNet discussed their use of on-demand talent in facing geospatial mapping challenges. Todd Bacastow from Maxar, Ryan Lewis from In-Q-Tel's CosmiQ Works lab, and Adam Van Etten from In-Q-Tel's CosmiQ Works lab sat down to discuss how SpaceNet has been able to use Topcoder to deliver value to their geospatial mapping projects.
If You Don’t Have a Community, Assemble One
SpaceNet (made up of members from In-Q-Tel's CosmiQ Works lab, Maxar, Intel AI, AWS, Capella Space, and Topcoder) uses machine learning for geospatial mapping that serves both governments and the commercial sector. Currently, geospatial mapping with satellite imagery is largely a manual process. With over 100 petabytes of satellite imagery archived, SpaceNet uses machine learning algorithms to decode satellite imagery and apply structure to all of the rich data that would otherwise gather dust in the archives.
It’s a big idea with big implications. When SpaceNet started, however, there wasn't a networked tech community to help build solutions. Topcoder’s community filled that gap. SpaceNet has run successful challenges that resulted in tangible algorithms. At the same time, they have leveraged Topcoder talent to help build broader, richer capabilities in this niche space.
Finding the Right Algorithm
With a small internal team, SpaceNet relies on talent communities like Topcoder to develop the algorithms they need to move their work forward successfully. With thousands of research papers in their area, SpaceNet tests a plethora of theories to see which algorithm models perform best at scale. And as image-capture technology evolves, Topcoder is able to help build new algorithms as fast as new tech comes in.
By leveraging Topcoder, SpaceNet can host challenges where each user approaches the project from a different viewpoint. This allows Topcoder's talent to flex their muscles and to intellectually explore a complex subject. And it gives SpaceNet the ability to build and test hundreds of algorithms without committing valuable internal staff time to test each one.
In addition to labeling projects, SpaceNet has envisioned multiple purposes for satellite imagery in the government and commercial spaces. One of these is the ability to map and label roads. This has a plethora of use cases — a few of which we talk about in this post. This challenge was tackled in SpaceNet 5. Check out the results here! Finally, you can view the complete TCO SpaceNet panel discussion here.