Browse Teach Curate

Slope-Aware Height Estimation for Satellite Terrain Multi-View Stereo

A learning-based multi-view stereo method for reconstructing large-scale Earth terrain from satellite imagery that injects terrain-slope information into height estimation. Students learn how a slope map derived from image gradients guides a refined height-interval partition at the micro level and a learnable Gaussian height-correction at the macro level, together with a slope-direction loss, to improve reconstruction accuracy over generic MVS networks.

Rosie Text

TS-SatMVSNet: Slope Aware Height Estimation for Large-Scale Earth Terrain Multi-view Stereo Song

Reconstructing 3D terrain from overlapping satellite images with learning-based multi-view stereo (MVS) usually reuses generic MVS networks that ignore how real terrain surfaces vary, limiting height…