The aim of this project was to use a convolutional-deconvolutional neural network to predict building height data from satellite images.

The team focused on Sao Paulo to build and train the model using three sources of satellite images. For phase 1, the team used high-resolution maxar data (not publicly available). For phase 2, the team explored the use of lower-resolution data from Sentinel 2, and radar images from Sentinel 1.

The full code for both phases can be found at Data for Children GitHub site.