GeoFormer achieves 3.19 m building-height RMSE with 0.32 M parameters by applying windowed local attention to Sentinel imagery, outperforming CNN baselines by 7.5 % while releasing all code and weights.
Refining urban morphology: An explainable machine learning method for estimating footprint-level building height,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
GeoFormer: A Lightweight Swin Transformer for Joint Building Height and Footprint Estimation from Sentinel Imagery
GeoFormer achieves 3.19 m building-height RMSE with 0.32 M parameters by applying windowed local attention to Sentinel imagery, outperforming CNN baselines by 7.5 % while releasing all code and weights.