Overview

One line of code to get embeddings from any Remote Sensing Foundation Model (RSFM) for any place and any time

Motivation¶

rs-embed background

The remote sensing community has seen an explosion of foundation models in recent years. Yet, using them in practice remains surprisingly painful. Model interfaces vary widely between imagery models and precomputed embedding products, input semantics are often ambiguous, and temporal, spectral, and spatial assumptions differ enough that cross-model comparison becomes tedious. In practice, even simple benchmarking can turn into glue-code work.

RS-Embed aims to fix this.

Goal

Provide a minimal, unified, and stable API that turns diverse RS foundation models into a simple ROI → embedding service — so researchers can focus on downstream tasks, benchmarking, and analysis, not glue code.

Start Here¶

Use the documentation the same way you would approach a mature library: start with one successful run, choose a model only after you understand the core API shape, and drop into the reference when you need exact contracts.

Start here if you want to...	Page
get something running quickly	Quickstart: install the package, run a first example, and learn the three core APIs
understand settings first	Before You Start: learn which settings change embedding quality, semantics, and runtime before you tune anything
choose a model	Models: shortlist model IDs by task, input type, and temporal behavior
check exact signatures	API: exact signatures for specs, embedding, export, and inspection
understand internals	Architecture: module map, call flows, and registration mechanics
add support for a new model	Extending: add a new model adapter or integrate with the registry/export flow

Why rs-embed¶

rs-embed is designed to make embedding acquisition work the same way whether you are probing one location once or building a large benchmark dataset. The public surface stays small enough for quick use, but still exposes the controls you need for fetch policy, output shape, export configuration, and model-specific options.

Most workflows revolve around three functions: get_embedding(...) for one ROI, get_embeddings_batch(...) for repeated inference with one model, and export_batch(...) for dataset or benchmark generation. The rest of the documentation expands on those paths rather than introducing a second parallel abstraction layer.

Common Tasks¶

Goal	Page	Main API
Get one embedding for one ROI	Quickstart	`get_embedding(...)`
Compute embeddings for many ROIs	Quickstart	`get_embeddings_batch(...)`
Build an export dataset	Quickstart	`export_batch(...)`
Compare model assumptions	Models	model tables + detail pages
Inspect a raw provider patch	Inspect API	`inspect_provider_patch(...)`

Advanced Reading¶

Once the basic flow is familiar, read Before You Start before changing model defaults for speed or quality. Then use Concepts for the semantics behind TemporalSpec, OutputSpec, and backends, Workflows for task-oriented recipes, and Advanced Model Reference for deeper preprocessing and temporal comparison details. Limitations collects current constraints and edge cases.

Citation¶

@article{ye2026modelplacetimeremote,
      title={Any Model, Any Place, Any Time: Get Remote Sensing Foundation Model Embeddings On Demand},
      author={Dingqi Ye and Daniel Kiv and Wei Hu and Jimeng Shi and Shaowen Wang},
      year={2026},
      eprint={2602.23678},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2602.23678},
}

License¶

This project is released under the Apache-2.0