Building-Damage-Detection-in-Ukraine

Predict the level of infrastructure damage in Ukraine using geospatial images and transfer learning techniques.

Methodology

A two pronged approach was use to address image segmentation and classification.

Data

To utilize with XView2 Data, please place the unzipped folder "train", "test", and "hold" within the data directory. This zips can be downloaded from the xview challenge; the files of interest are the "Challenge training/holdout/test sets".

Expected Repo File Structure

| data
----| classification_images
----| holdout_images
----| testing_images
----| train
--------| images
--------| labels
--------| targets
----| test
--------| images
--------| labels
--------| targets
----| mariupol_explore.ipynb
----| preprocessing_classification_data.ipynb
----| sentinel.ipynb
----| visualize_data.ipynb
| src
----| cnn.ipynb ----| segmentation.ipynb