Embed2Scale

⚠️ April 24, 3 p.m. CEST — Embed2Scale and the Earth2Vec community are hosting a webinar: "Embedding Space Metrics & Constrained Fine‑Tuning" Linh Le, Data Science Consultant at IBM, will introduce Embedding Space Metrics Constrained Fine‑Tuning (ESM‑CFT) — an optimisation method in the TerraTorch framework designed to preserve the structural integrity of the latent space during supervised fine‑tuning. ✅ For researchers & practitioners working on embeddings, geospatial ML, and representation learning ✅ Technical deep dive + live discussion 🎟️ Register here: https://lnkd.in/e_i9yPYf EUSPA - EU Agency for the Space Programme

📢 Embed2Scale at IGARSS 2026 On 9 August 2026, Embed2Scale will host a tutorial at IGARSS 2026, titled “A Hands‑On Introduction to Benchmarking Neural Compression and Representation Learning for Earth Observation.” The tutorial introduces NeuCo‑Bench, an open benchmarking framework developed within Embed2Scale to enable fair and reproducible evaluation of neural compression and representation learning methods for Earth Observation. Participants will be introduced to the foundations of neural compression and to NeuCo‑Bench as an extensible benchmarking framework. They will also receive practical guidance on benchmarking state‑of‑the‑art foundation models. 🔗 Read more: https://lnkd.in/ePmJFvpU EUSPA - EU Agency for the Space Programme

Over the past two years, Embed2Scale has released open datasets supporting research in embedding models. 👉 Explore the datasets, reuse them, or connect with their authors to collaborate further: https://lnkd.in/eB6ekq7C EUSPA - EU Agency for the Space Programme

After a first event in March, Embed2Scale and the Earth2Vec community are teaming up again for a new technical webinar: "Embedding Space Metrics Constrained Fine-Tuning." This webinar introduces Embedding Space Metrics Constrained Fine-Tuning (ESM-CFT), an optimisation method in the TerraTorch framework designed to preserve the structural integrity of the latent space during supervised fine-tuning. 📅 April 24, at 3 p.m. CEST 🎟️ https://lnkd.in/e_i9yPYf EUSPA - EU Agency for the Space Programme

Jonas Hurst from the Institute for Geoinformatics, University of Münster, represented the Embed2Scale project at the first Geo-Embeddings Sprint, which took place in early March 2026. The sprint was co-hosted by the Cloud-Native Geospatial Forum (CNG), Planet, and the Clark Center for Geospatial Analytics. The sprint brought together researchers and industry practitioners working on Geospatial Foundation Models to collaboratively define best practices for storing and exchanging Earth Observation embeddings. 🌍 One of the key outcomes of this first meeting is https://geoembeddings.org/ - a central resource for gathering and documenting best practices. EUSPA - EU Agency for the Space Programme

After a first event in March, Embed2Scale and the Earth2Vec community are teaming up again for a new technical webinar: "Embedding Space Metrics Constrained Fine-Tuning." This webinar introduces Embedding Space Metrics Constrained Fine-Tuning (ESM-CFT), an optimisation method in the TerraTorch framework designed to preserve the structural integrity of the latent space during supervised fine-tuning. 📅 April 24, at 3 p.m. CEST 🎟️ https://lnkd.in/e_i9yPYf EUSPA - EU Agency for the Space Programme

Last Friday, March 27, Embed2Scale and the Earth2Vec community hosted the webinar “Visual Prompting for Geospatial Image Segmentation based on Embedding Maps.” The recording is now available. It introduces VP‑GIS, a framework that uses high‑dimensional embedding maps to guide zero‑shot segmentation models for Earth Observation data. 🎥 Watch the recording: https://lnkd.in/ekp5wwXd EUSPA - EU Agency for the Space Programme

🚀 Today’s the day! Embed2Scale and the Earth2Vec community are hosting the webinar “Visual Prompting for Geospatial Image Segmentation based on Embedding Maps” at 3 p.m. CET. The session focuses on Visual Prompting for Geospatial Image Segmentation (VP‑GIS) and includes a prototype demonstration, an explanation of how the approach works, and an outlook on future developments. 👉 Register here: https://lnkd.in/eZTFzw-z EUSPA - EU Agency for the Space Programme

AI for Earth Observation? Our Embed2Scale project released TerraCodec: an open-source family of neural compression models designed for optical Earth Observation data. With TerraCodec you can "go beyond" RGB, while achieving more efficient storage, transfer and sharing of satellite data.

We have something special planned. 🌎 🚀 The next two Earth2Vec meetings will be dedicated 𝘁𝗲𝗰𝗵𝗻𝗶𝗰𝗮𝗹 𝘄𝗲𝗯𝗶𝗻𝗮𝗿𝘀, co-hosted with Embed2Scale. Earth2Vec is a community initiative around accelerating 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗮𝗻𝗱 𝗮𝗽𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀 𝗼𝗳 𝗴𝗲𝗼𝘀𝗽𝗮𝘁𝗶𝗮𝗹 𝗲𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀. We kick things off on March 27 with Duc K. presenting research on leveraging Earth Observation embeddings for zero-shot geospatial image segmentation through visual prompting. Featuring a demo, deep-dive and discussion. 👉 Don't miss it and sign up here: https://lnkd.in/eZTFzw-z 💻 Earth2Vec Community: https://lnkd.in/gBn-KtTB Embed2Scale EUSPA - EU Agency for the Space Programme Romeo Kienzler Thomas Brunschwiler Michele Lazzarini Gabriele Cavallaro Jonas Hurst Damien ROBERT Olivier Hersperger Rikard V. Rocco Sedona Roya Shokoohi Corneliu Octavian Dumitru Kennedy Adriko #EarthObservation #GeoAI