Open Cities AI Challenge: Segmenting Buildings for Disaster Resilience

Open Cities AI Challenge -- concluding report

The Open Cities AI Challenge, put on by GFDRR with Azavea and DrivenData, recently concluded with over 1,100 participants, 2,100 submissions and $15,000 in total prizes awarded. Along two competition tracks, the Challenge produced global public goods — open-source data, code, research, and know-how — that will support mapping efforts for disaster resilience.

This includes:

Increasing availability and accessibility of ML-ready drone imagery and locally created geographic data, especially in the underrepresented areas of African cities, via the Open Cities AI Challenge dataset

Building footprint segmentation models that generalize well to different and unseen geographies and a range of drone imagery characteristics

Nuanced explorations into complex Responsible AI dimensions of developing and applying automated systems for disaster risk management

Take a look at the Open Cities AI Challenge concluding report for more information about the challenge, and check out winning entries from the Building Segmentation and Responsible AI tracks of the competition on GitHub and OpenDRI.