IARPA announces Functional Map of the World challenge winners
On March 16, the Intelligence Advanced Research Projects Activity, within the Office of the Director of National Intelligence, announced the winners of the functional Map of the World—“fMoW”—Challenge, a competition that called for contestants to develop automated techniques to accurately categorize points of interest from satellite imagery. There were 69 participants, “solvers,” who competed for a total $100,000 prize purse; the top three developers Open Sourced their solutions that can be used to further improve geospatial computer vision efforts. First through fifth place winners were comprised of both industry and academia.
The fMoW challenge launched in July 2017 and sought solvers from around the world to develop machine learning algorithms, or other automated techniques, to accurately categorize points of interest from satellite imagery. To encourage the development of innovative solutions, IARPA released one of the largest annotated satellite imagery datasets for participants to use to train, validate, and test their solutions throughout the challenge.
“The dataset itself has huge value. We are expecting people to make future innovations using this dataset, and now having three open source solutions adds additional value by allowing us to benchmark what is possible today. People can apply these open source algorithms to develop better algorithms in the future,” said Dr. HakJae Kim, IARPA program manager. “I was pleased with the complexity and diverse range of novel solutions produced by challenge competitors. These finalists have all made significant contributions to fostering innovation through crowd-sourcing and together they formed a collaborative community that will help to advance new and better methods of automated image classification and machine intelligence technologies.”
The top solvers presented their winning solutions during a government workshop in Washington, D.C., on March 14, 2018. Following the workshop, IARPA released a large portion of the ground truth and test data set. The data together with the open source solutions, can be found on the challenge microsite. To learn more about the challenge please visit https://www.iarpa.gov/challenges/fmow.html.