Related packages

TerraTorch is part of a larger open-source ecosystem for geospatial AI.

Frameworks

TerraTorch uses Lightning as a training and inference engine. Lightning handles many basic tasks like GPU allocation, logging and more. It uses PyTorch as the machine learning framework.

The tasks and data modules in TerraTorch are based on TorchGeo. Therefore, TorchGeo datasets are directly compatible with TerraTorch.

Data Tools

Xarray is a well suited tool for handling multidimensional data and supports lazy loading and more.
It's extension rioxarray supports reading or writing tif files and handles geospatial functionalities like the CRS.

For geospatial table data like polygons, GeoPandas is a well suited tool.

Model Libraries

TerraTorch includes meta registries for backbones, decoders and more. Models from Pytorch Image Models (timm) and Segmentation Models PyTorch (SMP) are directly available in TerraTorch.