Search papers, labs, and topics across Lattice.
GeoCore-9B is a novel 9-billion-parameter generative foundation model specifically trained from scratch on Earth observation (EO) data, overcoming the limitations of existing models that rely on natural image priors. Utilizing a Flow Matching-based Diffusion Transformer and a Geospatial Semantic Alignment loss, it effectively integrates text descriptions and geospatial metadata to enhance generation quality. The model demonstrates superior performance in visual fidelity and geographic structural accuracy, achieving state-of-the-art results in complex EO tasks like cloud removal and SAR-to-optical translation.
GeoCore-9B sets a new benchmark in Earth observation generative modeling by achieving unprecedented accuracy and versatility through geospatially aware training.
Existing generative models for earth observation (EO) predominantly rely on fine-tuning natural image priors, which limits their scalability and introduces perspective biases that conflict with geospatial constraints. To address this, we introduce GeoCore-9B, a 9-billion-parameter generative foundation model, which is the first of its scale to be trained from scratch exclusively on EO data. Unlike previous EO foundation models, GeoCore-9B is built upon a Flow Matching-based Diffusion Transformer (DiT) and natively conditions generation on text descriptions and continuous geospatial metadata, including ground sample distances, latitudes, and longitudes. To overcome the convergence and spatial disorientation challenges of training at this scale, we propose a Geospatial Semantic Alignment loss. This objective distills structural Earth surface priors (e.g., terrain and urban areas) from a frozen specialist teacher network, constraining the diffusion latent trajectory during training without adding inference overhead. Pre-trained on the global-scale Git-10M dataset, GeoCore-9B demonstrates strong downstream versatility. Beyond standard proxy generative tasks, we show that GeoCore-9B can be effectively adapted for practical EO applications, including highly challenging tasks such as cloud removal and SAR-to-optical cross-modal translation. Extensive evaluations confirm that GeoCore-9B establishes new state-of-the-art performance in both visual fidelity and geographic structural accuracy.
