Inspiring individuals and organizations to develop state-of-the-art machine learning algorithms that advance computer vision capabilities to address global challenges.
Taking computer vision to the next level one challenge at a time.
Computer vision technology has the potential to help solve global environmental and humanitarian challenges. DIU has partnered with Department of Defense organizations, federal, state, and local first responders, and non-governmental organizations to run a series of xView competitions with a variety of real-world applications.
The competitions benchmark machine learning algorithms and stimulate innovative ideas to overcome analytical roadblocks and make the growing supply of aerial and satellite imagery actionable. Winning xView algorithms are available to international communities of interest and have been put to global use, for example, in the aftermath of hurricanes and wildfires.
Challenge has started
xView3 Challenge: Detecting Illegal, Unreported, and Unregulated Fishing Vessels
Illegal, unreported, and unregulated (IUU) fishing is a major threat to human food supply, marine ecosystem health, and geopolitical stability. The effects of this exploitation are extensive: coastal environments and social systems are destabilized and maritime developing countries suffer acute economic losses, creating conditions for increased crime and human rights abuses. Moreover, IUU fishing exacerbates the impact of climate change and increases the likelihood of military conflict over scarce ocean resources.
Rapid detection of IUU fishing activity would enable earlier interdiction and prosecution of offenders, mitigating these damages and reestablishing regional stability. The rise of synthetic aperture radar (SAR) satellite imagery offers an all-weather tool to detect vessels that may otherwise elude fishery enforcement authorities. xView3 challenges competitors to develop algorithms that identify and classify SAR satellite images of vessels suspected of engaging in IUU fishing.
COMPETITION DETAILS
The xView3 challenge was organized and run by DIU in collaboration with Global Fishing Watch (GFW). During the three-month competition, 1,900 registrants from 67 countries created machine learning (ML) models to identify and differentiate between fixed infrastructure, fishing vessels, and non-fishing vessels in one of the largest open-source datasets of all-weather, day-and-night maritime radar imagery.
The xView3 Challenge offered two tracks for competition: one for open source algorithms and the other for closed source algorithms. DIU and GFW awarded $150,000 in total prizes. The winning model showed a substantial improvement, attaining an aggregate score three times higher than the government reference model released in the Challenge’s dataset. xView3 finalist details were shared here. Competition terms and conditions are detailed here.
PARTNERS
Global Fishing Watch
A leading international nonprofit organization dedicated to advancing ocean governance through increased transparency of human activity at sea.
National Oceanic and Atmospheric Administration
A scientific agency in the U.S. Department of Commerce that focuses on the conditions of the oceans, major waterways, and the atmosphere.
National Maritime Intelligence-Integration Office
NMIO integrates maritime intelligence, improves information sharing, and fosters domain awareness to protect the United States, its allies, and partners against threats to, from, and in the global maritime domain.
U.S. Coast Guard
One of the primary U.S. organizations responsible for combating IUU fishing, often in collaboration with international partners.
COMMON QUESTIONS
What happens when I pre-register for xView3?
You will receive email updates about the challenge start date, and, when available, further instructions related to the competition.
What happens to the winning algorithms?
Open source algorithms will be made freely available to both government and non-governmental organizations combating IUU fishing.
DIU prize recipients will be eligible (but not guaranteed) for follow-on Other Transaction agreements under IAW 10 U.S.C. § 2374 Prize Challenge Authority.
What is synthetic aperture radar (SAR)?
Synthetic-aperture radar (SAR) is a type of radar. Radar creates images by transmitting microwave radiation and reconstructing signals that are returned via reflection or backscatter. A major difference between SAR and traditional radar is that SAR can improve image resolution by using the rapid movement of the antenna (e.g. on a satellite) itself to simulate a much larger antenna.
xView3 Challenge Latest:
xView2 Challenge: Building Damage Assessment
The time it takes to properly assess damage in the wake of a major event can be the difference between life and death. However, emergency responders must often navigate disruptions to local communication and transportation infrastructure, making accurate assessments dangerous, difficult, and slow. While satellite and aerial imagery offer a less risky alternative that covers more ground, analysts must still conduct manual, time-intensive assessments of images.
The xView2 Challenge used high-resolution RGB satellite imagery and competitors developed algorithms to identify buildings and score structural damage before and after a variety of natural disasters around the world.
Winning models were deployed in various disaster relief efforts and data remain available for download. Visit the xView2 website to learn more.
THE RESULTS
The competition resulted in more than 2,000 submissions and multiple winning solutions with an 80% damage assessment success rate (the first-place algorithm was 266% more accurate than the government baseline). The top solutions have since been deployed to assist with the 2020 California wildfires, coastal hurricanes, and the 2019-2020 Australia Bushfires. xView2 has also yielded substantial positive engagement with the AI community via open-source releases of model code and a machine learning dataset called xBD that continues to be a resource to the HADR community.
xView2 satellite imagery assessment of a location (right) identifies damaged structures after California's 2020 Valley Fire.
End Users
Federal Emergency Management Agency
National Geospatial-Intelligence Agency
National Aeronautics and Space Administration
California Department of Forestry and Fire Protection
California Air National Guard
California Governor’s Office of Emergency Services
The World Bank
The United Nations Satellite Centre UNOSAT
PARTNERS
California Governor’s Office of Emergency Services
National Security Innovation Network
Federal Emergency Management Agency
National Aeronautics and Space Administration
United States Geological Survey
California Department of Forestry and Fire Protection
Joint Artificial Intelligence Center
California Air National Guard
Carnegie Mellon University, Software Engineering Institute
National Geospatial-Intelligence Agency
xView2 Challenge Latest:
xView1 Challenge: Object Detection
The first xView Challenge was completed in 2018 and focused on detecting a wide variety of objects in high-resolution, commercial satellite imagery. Competitors’ algorithms had to identify objects as they actually appear in the world, with realistic occurrence frequency (common vs. rare), size ranges (cars vs. airports), and fields of view.
Notably, the xView1 Challenge marked the release of the largest and most diverse publicly available dataset of overhead imagery used to assess competitors’ algorithms. The dataset contains more than one million labeled bounding box annotations across 60 classes, covers 1,415 square kilometers of complex scenes spanning the globe, and sets a new standard for objects that are small, rare, fine-grained, and rigid.
The challenge has finished but data is still available for download. Visit the xView1 website to learn more.
THE RESULTS
The competition resulted in over 2,000 submissions, top algorithms outperformed government baseline by over 200% and laid the groundwork to launch the xView2 Challenge the following year.
PARTNERS
National Geospatial-Intelligence Agency (NGA)