EarthquakeChangeDetection

This repository aims to assess the damage caused by the devastating earthquakes that hit Turkey and Syria on Feb 6, 2023. Here, we provide pre & post disaster satellite images acquired from Planet, as well as damaged buildings detected using Chooch AI Inference Engine (www.chooch.com).

Directory Structure

Pre & post disaster satellite images spatially matched are provided under ./data_matched/ directory. The files_info.csv contains useful information and paths to image pairs.

The image pairs are grouped by city and district such that each pair can be accessed as:

{city}/{district}_{id}/
-- {...}_pre.jpg
-- {...}_post.jpg
-- {...}_pre.tif
-- {...}_post.tif
-- {...}_result.jpg
-- {...}_result.json

.tif files contain geospatial information. .tif files and .jpg files have the same image content.
For some pairs, our preliminary results that compare pre&post images are provided as _results.jpg and _results.json.

Compared Objects Format

Result JSON

Each json has two components, raw_results and changes, where

• raw_results contain the original predictions for 'A' & 'B' images and
• changes contain the predictions grouped by class name and belonging to 'only_A','only_B','common_A','common_B'

where in this case A refers to pre-disaster and B refers to post-disaster buildings. The objects listed under 'only_A' are destroyed buildings.

Comparison Image

The images are shown side by side (or top/bottom, depending on aspect ratio) and the bounding boxes are drawn in two colors, such that:

• blue for common (undamaged or minor damage) buildings
• red for objects that are present only in pre disaster image, therefore buildings with major damage or destroyed.