Welcome to TerraTorch#


The Geospatial Foundation Models Toolkit
-
Fine-tuning made easy
Run
pip install terratorch
and get started in minutes. -
Overview
Get an overview of the functionality and architecture with in-depth explanations.
-
The big picture
We outline the ideas and reasons why we build TerraTorch in the paper.
-
Open Source
TerraTorch is distributed under the terms of the Apache 2.0 license.
The purpose of this package is to build a flexible fine-tuning framework for Geospatial Foundation Models (GFMs) based on TorchGeo and Lightning which can be employed at different abstraction levels. It supports models from the Prithvi, TerraMind, and Granite series as well as models from TorchGeo and timm.
This library provides:
- Ease-of-use and all the functionality from Lightning and TorchGeo.
- A modular model factory that combines any backbone with different decoders for full flexibility.
- Ready-to-use tasks for image segmentation, pixelwise regression, classification, and more.
- Multiple abstraction levels and inference pipelines to power enterprise applications.
A good starting place is familiarization with PyTorch Lightning, which this project is built on. TorchGeo is also an important complementary reference with many available models and specific datasets. If you have any open questions, please check the FAQs or open an issue in GitHub.