AI4EO Spring School 2026

Day 1

Presentation by Iris de Gélis

Iris presented her PhD research on change detection in 3D point clouds using deep learning. Her talk was well-structured and accessible. I found her presentation balanced between technical depth and clarity which made it engaging for both specialists and novices to the topic.

Presentation by Stéphane May

Stéphane delivered a comprehensive introduction to Foundation Models. Althought this talk was intended as an introduction for attendees less familiar with the field, like myself, some sections remained difficult to follow.

Practical Session by Pierre Adorni

The practical walked participants through flood mapping using Sentinel-1 radar and Sentinel-2 optical imagery. It was a good hands-on illustration of how foundation models perform on real Earth observation tasks. There was a technical hiccup mid-session but it was quickly handled by Pierre. If anything, this only reinforced how capable and knowledgeable he is. That said, meeting Pierre was one of the personal highlights of the spring school, sharing a drink with him after a long day of sessions was a genuine pleasure.

Day2

Presentation by Pedram Ghamisi & Weikang Yu

This talk genuinely stood out for me. It brought an ethical lens to AI and geospatial data and how AI must be developed and deployed ethically within the field. The talk covered bias mitigation, AI security, geo-privacy, and the broader societal implications of using AI-driven satellite data. I found both the depth of Pedram Ghamisi's and Weikang Yu's research impressive. The questions they are raising for our community are relevant and important for consideration.

Presentation by Adam Stewart

Adam Stewart presented TorchGeo, a Python library designed to make it easier for machine learning practitioners to work with geospatial and remote sensing data within the PyTorch ecosystem.

Day3

Presentation by Nicolas Audebert

The talk and practical session by Nicolas Audebert on Conditional Flow Matching was a very nice. It was clearly explained, and a lot of fun to work through.

Presentation by ESA Phi-lab

The ESA Phi-Lab session showcased the real-world challenges they are tackling at the intersection of AI and satellite Earth Observation, and highlighted opportunities for the research community to engage with their work. Although the session was held remotely, it was very nice to see the faces behind this research group. Finally the conference closed with the announcement of their "Reaching New Heights with GeoFM" challenge.