Detailed Glacier Area Change Analysis in the European Alps with Deep Learning

@NeurIPS 2023 Workshop on Tackling Climate Change with Machine Learning

For the paper and presentation please check the project page.

Set up the python environment

    conda env create -f environment.yml
    conda activate glsegenv
    cd gl_seg_dl

Reproduce the results

Data downloading:

1. Download the glacier outlines: bash ./scripts/download_outlines.sh
2. Download the glacier rasters (stored here: https://huggingface.co/datasets/dcodrut/glacier_mapping_alps): bash ./scripts/download_data.sh
   The archived rasters have ~3Gb for each year (inventory one & 2023). After extracting the NetCDF rasters, we will need 20Gb for each year.

Data processing:

1. Cross-validation splits & patch sampling: python main_data_prep.py
2. Compute the mean & stddev of the training patches: python main_compute_data_stats.py

Model training, testing and area estimation:

1. Train the five models: bash scripts/train.sh (by default it runs on a single GPU; check the bash script for running on multiple GPUs)
2. Apply model on each glacier both on the inventory images and the 2023 ones: bash scripts/infer.sh
3. Estimate the areas based on the predictions from the previous step: bash scripts/eval.sh