Ritwik Gupta and Carson Sestili describe their use of deep learning in IARPA’s Functional Map of the World Challenge.

“The idea is how can you take these very minute differences, not only in scale, but also in landscape, the buildings on there, etc., and identify different land functions. This makes it very different from just a traditional image classification problem because you have to take in not only the object of interest, which is like let’s say a building, but also its entire surroundings.”