Search papers, labs, and topics across Lattice.
Department of Computer Science and Technology, University of Cambridge, Cambridge, UK
3
0
4
3
Geospatial foundation models seem to handily beat standard satellite features on remote sensing benchmarks鈥攗ntil you enforce a 10 km spatial buffer, which virtually erases their statistical advantage.
Pretraining loss is a misleading metric, leading to inefficient compute use, while a simple scaling rule can unlock superior performance in pixel-wise Earth-observation models.