GeoCore-9B: New generative model for Earth observation trained on geospatial data

Researchers have introduced GeoCore-9B, a new 9-billion-parameter generative foundation model specifically designed for Earth observation tasks. Unlike previous models that fine-tuned natural image priors, GeoCore-9B is trained exclusively on Earth observation data using a Flow Matching-based Diffusion Transformer. It natively incorporates geospatial metadata such as latitude, longitude, and ground sample distances into its generation process. To improve training stability and accuracy, a Geospatial Semantic Alignment loss was developed to distill structural Earth surface priors from a specialist teacher network. AI

IMPACT Establishes new state-of-the-art for Earth observation tasks like cloud removal and SAR-to-optical translation.

RANK_REASON The cluster describes a new research paper detailing a novel AI model for a specific domain. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• Diffusion Transformer
• Earth observation
• Flow Matching for Generative Modeling
• GeoCore-9B
• Geospatial Semantic Alignment
• Git-10M
• synthetic aperture radar

• model release
• paper

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →