Source: Deep Learning on Medium
Announcing SpaceNet 6: Multi-Sensor All Weather Mapping
Preface: SpaceNet LLC is a nonprofit organization dedicated to accelerating open source, artificial intelligence applied research for geospatial applications, specifically foundational mapping (i.e. building footprint & road network detection). SpaceNet is run in collaboration with CosmiQ Works, Maxar Technologies, Intel AI, Amazon Web Services (AWS), Capella Space, Topcoder, and IEEE GRSS.
Spanning three years, featuring five unique datasets and challenges, SpaceNet has continued to focus on different aspects of applying machine learning to solve difficult foundational mapping problems. SpaceNet has open sourced ~27,000 sq. km of imagery, >800,000 building footprints, and ~20,000 km of road labels across 10 cities. With these datasets, our challenges have focused on several different geospatial aspects including high resolution building footprint mapping, the effects look angle on model performance, and building routable road networks. As we look ahead to our next challenge: SpaceNet 6, we will push into a new frontier and an under-explored modality of data: Synthetic Aperture Radar (SAR).
A New Modality — The SpaceNet 6 Dataset
Synthetic Aperture Radar sensors are unique as they can penetrate clouds and collect during all weather conditions. Furthermore, radar satellites do not require illumination and can capture data during both the day and the night. Consequently, overhead collects from SAR satellites could be particularly valuable in the quest to aid disaster response, when weather and cloud cover can obstruct traditional electro-optical sensors.
Despite all of these advantages, there is little open data available to researchers to explore the effectiveness of SAR for such applications, particularly at ultra-high resolutions. Thanks to our new SpaceNet partners Capella Space, SpaceNet will be the first to feature a new open-source dataset of half-meter SAR in a challenge setting. Additionally, to compliment the SAR collect, the dataset will also feature half-meter electro-optical imagery from Maxar’s WorldView 2 satellite. Our area of interest for this challenge will be centered over the largest port in Europe: Rotterdam, the Netherlands. Such an area features thousands of buildings, vehicles, and boats of various sizes, which will make for an effective test bed for SAR and the fusion of these two types of data.
Competition Structure and CVPR EarthVision Workshop
In the SpaceNet 6 challenge, participants will be asked to automatically extract building footprints with computer vision and artificial intelligence (AI) algorithms using a combination of these two diverse remote sensing datasets. For training data, participants will be allowed to leverage both the electro-optical and SAR datasets. However, for scoring performance, only a subset of data will be available to participants to ultimately map buildings. We hope that such a structure will incentivize new data fusion methods and other approaches such as domain adaptation.
The SpaceNet 6 challenge will run for approximately two months, with an anticipated launch date of March, 2020. Data preprocessing is ongoing, and the dataset will be publicly released when the competitions nears. Evaluation will again rely on the SpaceNet metric, implementing the F1 score, which represents the harmonic average of precision and recall.
Finally, following the challenge, some of SpaceNet’s best participants will be invited to share their work at one of CVPR’s most respected workshops: EarthVision 2020. Look out for more details on how you can participate and take part in the workshop.
 Dukai, B. (Balázs) (2018) 3D Registration of Buildings and Addresses (BAG). 4TU.Centre for Research Data. Dataset. https://doi.org/10.4121/uuid:f1f9759d-024a-492a-b821-07014dd6131c