This repository contains scripts for downloading Sentinel-2 satellite imagery and pre-processing it for the ibm-nasa-geospatial/Prithvi-100M-multi-temporal-crop-classification model.

It also contains an Apptainer container recipe providing a software environment for running the model locally.

python download_images.py coords.json

coords.json is a mandatory positional argument. It should be a JSON document where each key represents a location name, and its corresponding value is a list of latitude/longitude pairs that define the desired polygon for that location.

(cd <input_dir>; unzip \*.zip)
python stacker.py <input_dir> [<value_divisor>]

value_divisor is an optional argument that divides the input pixel values, which may improve model performance. It defaults to 1, which does not change the pixel values.

Upload the .tif file to the model's demo space and click submit.