Awesome

cogsima2022

Repository for the paper Enhancing Land Subsidence Awareness via InSAR Data and Deep Transformers

input	output

Installation

Clone and install the dependencies

git clone https://github.com/galatolofederico/cogsima2022.git
virtualenv --python=python3.8 env && . ./env/bin/activate
pip install -r requirements.txt

Download dataset

Download the shapefiles

wget http://131.114.50.176/owncloud/s/66EveoWWyvxd9BQ/download -O ./dataset.zip
unzip dataset.zip

Download and unzip DEM-v1.1-E40N20 in ./dataset/dem from copernicus

Build the raster (it will require some time and at least 32G of ram)

./build-raster.sh

If you want to re-split the dataset run

python -m scripts.split-dataset --data-folders ./dataset/raster/bologna-asc/ ./dataset/raster/bologna-dsc/ ./dataset/raster/pistoia-asc/ ./dataset/raster/pistoia-dsc/

Download pre-trained models

To download the pre-trained models run

wget http://131.114.50.176/owncloud/s/C0XJcCLAps0513s/download -O ./models.zip
unzip models.zip

Training

To train all the models run

./train-all.sh

To train a specific model run

python train.py --model <model> --train-batches 10000 --save

where model can be encoderencoder vitencoder encoderdecoder vitdecoder

Evaluation

To run the inference on the testing set on all the models run

./predict-all.sh

To run the inference on the testing set on a specific model run

python predict.py --model <model-path> --points <input-points> --eval-batches 1000

To compute all the metrics and plots from the paper run

python evaluate.py

Results will be available in ./results

Prediction

To run the regression on all the missing data in a shapefile run

./predict-fill-shp.sh -m <model> -s <input-shapefile> -f <field-name> -o <output-shapefile> -n <montecarlo-steps>

Contributions and license

The code is released as Free Software under the GNU/GPLv3 license. Copying, adapting and republishing it is not only allowed but also encouraged.

For any further question feel free to reach me at federico.galatolo@ing.unipi.it or on Telegram @galatolo