Hurricane Spatial Damage Assessment using Convolutional Neural Networks and High-Resolution Satellite Imagery.
CS 229 Final Project by Lukas Haas and Michal Skreta.
This is our main work performed on Google Cloud. It contains the data loading, augmentations, model training, evaluation, and interpretability components.
This file contains some data preprocessing utilities we built. In particular, in contains tools to visualize the damage overlays, match satellite images with their corresponding labels, cut out buildings from images, and resizing them to required sizes.
This file contains data preparation to upload it to Google Cloud as well as some local experimentations we performed on the data.