TyphoonGAN or G-DANS (Generative Destructive Adversarial Networks), is a novel framework to rapidly simulate typhoon aftermath and predict destruction hotspots based on satellite imagery, assisting disaster relief efforts. Built upon CycleGAN, it could be applied to areas where typhoon had never been an occurence.
We showcased G-DANs in the HADR Challenge organized by SSTA, Airbus and (DSTA), and in the NAPROCK international contest, clinching 1st runner up and special prize respectively.
Links
Technologies
GANs
Deep Learning
AI