A multimodal satellite foundation model trained with contrastive learning, modality-aware patch embeddings, cross-attention fusion, and a dual-centering regularizer achieves state-of-the-art results on GEO-Bench and Copernicus-Bench.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
TerraFM: A Scalable Foundation Model for Unified Multisensor Earth Observation
A multimodal satellite foundation model trained with contrastive learning, modality-aware patch embeddings, cross-attention fusion, and a dual-centering regularizer achieves state-of-the-art results on GEO-Bench and Copernicus-Bench.