Embed2Scale session @AI4Good: Embedding workflows for Earth Observation tasks

This workshop, with Embed2Scale expert Isabelle Wittmann from IBM Research, will demonstrate how modern Foundation Models can generate EO embeddings that preserve task-relevant information. These embeddings allow lightweight decoders up to two orders of magnitude smaller, accelerating training and inference.

In two segments, participants will:
– Gain hands-on experience with TerraTorch, an open-source library offering tools to extract embeddings via a no-code CLI, adjust aggregation levels for different use cases, visualize and analyse embedding properties and run complete workflows for downstream tasks.
– Be introduced to NeuCo-Bench, a benchmarking framework from the Embed2Scale consortium. NeuCo-Bench enables rapid evaluation of embeddings across diverse tasks within seconds, requiring no prior setup.

The workshop will conclude by integrating TerraTorch and NeuCo-Bench in a full workflow, showcasing instant evaluation of generated embeddings. It highlights efficiency gains, demonstrates practical workflows, and shows how embeddings drastically reduce computation and training time, with interactive demos ensuring participants leave with actionable skills for EO tasks.

By the end of this session, participants will be able to:
– Get an introduction to neural embeddings and their role in modern EO workflows.
– Learn to extract EO embeddings for your data using TerraTorch’s no-/low-code CLI.
– Benchmark embedding methods within seconds using NeuCo-Bench.
– Build your own embedding downstream task workflows and integrate NeuCo-Bench evaluations into them.