CS229 Hurricane Spatial Damage Assessment

Hurricane Spatial Damage Assessment using Convolutional Neural Networks and High-Resolution Satellite Imagery.

CS 229 Final Project by Lukas Haas and Michal Skreta.

Contents

Cloud Training, Evaluation, & Interpretability

This is our main work performed on Google Cloud. It contains the data loading, augmentations, model training, evaluation, and interpretability components.

Data Preprocessing Utilities

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.

Data Preparation & Experimentation

This file contains data preparation to upload it to Google Cloud as well as some local experimentations we performed on the data.