Cross-modal Sentinel-2 Image Retrieval with Global Imagery Embeddings

Recent advancements in Earth embeddings have enabled efficient analysis of geospatial data. EarthEmbeddingExplorer is an open-source web application that democratizes access to Earth embeddings by integrating multiple state-of-the-art models, including SatCLIP, FarSLIP, and SigLIP, to support cross-modal retrieval of Sentinel-2 imagery through text, image, and geographic location queries.

The application leverages the MajorTOM Core-S2L2A dataset, with approximately 250,000 pre-computed embeddings per model based on uniform spatial sampling, providing representative global coverage of 1.2% of Earth's land surface. All models and datasets are hosted on open-source frameworks (ModelScope and Hugging Face) to ensure accessibility.The interface enables intuitive visualization of geographical distribution, rendering top-match thumbnails, and exporting metadata. The web applications are accessible at Hugging Face Space.