Researchers have developed GeoCR, a novel generalist model designed for cloud removal in satellite imagery. Unlike previous methods that are dataset-specific, GeoCR can handle heterogeneous observations across different sensors, spectral bands, and temporal settings, even incorporating synthetic aperture radar (SAR) guidance. The model achieves this by using a shared latent interface that connects a pretrained RGB autoencoder with a flow transformer, allowing it to jointly process various data types. GeoCR was pre-trained on over 880,000 cloud-free images from ten datasets, demonstrating its ability to learn a universal cloud removal prior and perform effectively without dataset-specific fine-tuning. AI
IMPACT This generalist model could streamline cloud removal processes for satellite imagery analysis across various datasets and sensor types.
RANK_REASON The item describes a new research paper detailing a novel model for a specific task (cloud removal in satellite imagery). [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
