We are delighted to announce that our team will present two papers at the IEEE/CVF International Conference on Computer Vision (ICCV) 2025, taking place in Honolulu, Hawaii (October 19–23, 2025). Even more excitingly, our paper “Towards a Unified Copernicus Foundation Model for Earth Vision” has been shortlisted for the Best Paper Award at ICCV 2025! 

These works tackle key challenges in AI for Earth Observation (AI4EO) — how to build foundation models that truly understand our planet from space, and how to make these models trustworthy when predicting changes in our environment. The works were developed within the framework of the EU project ThinkingEarth, which brings together leading European research institutions and industry partners — including the Technical University of Munich, the National Observatory of Athens, and NVIDIA — to advance trustworthy, large-scale, and sustainable Earth intelligence through artificial intelligence and foundation models.
 

Towards a Unified Copernicus Foundation Model for Earth Vision
Led by Yi Wang (Technical University of Munich)
We introduce Copernicus-FM, a unified foundation model trained on 18.7 million images from all major Copernicus Sentinel missions. It can process any spectral or non-spectral data — from land to atmosphere — bridging the silos between Earth observation, weather, and climate research.
Read more

On the Generalization of Representation Uncertainty in Earth Observation
Led by Spyros Kondylatos and Nikolaos Ioannis Bountos (National Observatory of Athens)
This paper explores how AI models can quantify their own uncertainty when analyzing satellite data. It pioneers representation uncertainty in EO, showing that models pretrained on EO data generalize uncertainty far better than those trained on natural images — a key step toward trustworthy AI for Earth observation.
Read more

Congratulations to all collaborators for this outstanding achievement and for making it to the final list of ICCV Best Paper Award candidates!