IMPLICITY: CITY MODELING FROM SATELLITE IMAGES WITH DEEP IMPLICIT OCCUPANCY FIELDS

Abstract. High-resolution optical satellite sensors, combined with dense stereo algorithms, have made it possible to reconstruct 3D city models from space. However, these are, in practice, rather noisy and tend miss small geometric features that are clearly visible the images. We argue one reason for limited quality may be a too early, heuristic reduction of triangulated point cloud an explicit height field or surface mesh. To make full use underlying images, we introduce IMPLICITY, neural representation scene as implicit, continuous occupancy field, driven by learned embeddings pair ortho-photos. show this enables extraction high-quality DSMs: image resolution 0.5 m, IMPLICITY reaches median error ≈0.7m outperforms competing methods, especially w.r.t. building reconstruction, featuring intricate roof details, smooth surfaces, straight, regular outlines.