---
title: README
emoji: 🏃
colorFrom: purple
colorTo: red
sdk: static
pinned: false
---

## IBM and ESA 🌍 Advancing Geospatial AI with TerraMind & MS-CLIP

IBM and the European Space Agency (ESA) Φ-lab introduce TerraMind, the first multimodal, any-to-any generative foundation model for Earth Observation. 
TerraMind reaches new SOTA results on downstream tasks and introduces the "Thinking-in-Modalities" approach to handle missing modalities.
The model was pre-trained in partnership with Jülich Supercomputing Centre (Forschungszentrum Jülich) and supported by the FAST-EO project.

###  🎉  TerraMind has been accepted at ICCV 2025
We’re honored to present our work at one of the most prestigious conferences in computer vision. Stay tuned for more details!

###  🎉  MS-CLIP has been accepted at ECML 2025
We’re honored to present our work at one of the most prestigious conferences on machine learning. Stay tuned for more details!

###  🎉  TerraMesh has been accepted at CVPR EarthVision 2025
We’re honored to present our work at one of the most well-known workshops for Earth observation. Stay tuned for more details!


###  🤝 Open Source

The FAST-EO partners and ESA Φ-lab are committed to Open Source, sharing knowledge, and driving innovation for the geospatial domain.

###  📚 Papers and Resources

Explore the technical paper detailing TerraMind's architecture and applications [here](https://arxiv.org/abs/2504.11171). Details on MS-CLIP can be found [here](https://arxiv.org/pdf/2503.15969).

###  📊 Upcoming Models and Datasets

Stay tuned for our upcoming model and dataset releases like [TerraMesh](https://arxiv.org/abs/2504.11172).

###  🚀 Challenge

Already working with TerraMind? Submit your use case to the [TerraMind Blue-Sky Challenge](https://huggingface.co/spaces/ibm-esa-geospatial/challenge), a bi-monthly award spotlighting the boldest, most imaginative ways using TerraMind.


Explore our models, datasets, and papers, and join our discussions to shape the future of geospatial AI!