rs-embed

rs-embed is a unified Python toolkit specifically designed to simplify the process of obtaining embeddings from various Remote Sensing Foundation Models (RSFMs). It addresses the common challenges faced by researchers, such as inconsistent model interfaces, ambiguous input semantics, and diverse temporal, spectral, and spatial requirements. By providing a minimal, unified, and stable API, rs-embed aims to transform complex RSFMs into a simple region-of-interest (ROI) to embedding service. This approach allows users to focus on downstream tasks, benchmarking, and analysis rather than the intricacies of integrating different models.

The toolkit offers a unified interface for both on-the-fly models and precomputed products, ensuring consistency across diverse embedding sources. Key features include spatial and temporal specifications that describe desired outputs rather than requiring detailed fetching instructions. It also prioritizes batch export as a first-class workflow, facilitating large-scale experiments and data preparation. Compatibility wrappers are preserved, offering flexibility for advanced users while maintaining a clear learning path.

The motivation behind rs-embed stems from the rapid proliferation of foundation models in remote sensing and the practical difficulties in deploying them. By standardizing access and output, the toolkit enables easier comparison of multiple models within a single experiment, fostering more efficient research and development in the geospatial community.