LNEM: Lunar Neural Elevation Model
Suwan Lee (KENTECH), Jo Ryeong Yim (KARI), Kibaek Park (KENTECH), Dong Gyu Kim (KARI), Eunhyeuk Kim (KARI), Minsup Jeong (KASI), Chae Kyung Sim (KASI), Seokju Lee (KENTECH)
IEEE/CVF Conference on Computer Vision and Pattern RecognitionΒ (CVPR), 2026
We introduce LNEM, a neural elevation model for lunar terrain reconstruction that explicitly incorporates rigorous pushbroom sensor modeling to enforce multi-orbit consistency. By modeling pushbroom imaging geometry, LNEM reconstructs high-fidelity and geometrically consistent elevations from sparse multi-orbit imagery. We further present Lunar Studio, a standardized multi-orbit benchmark integrating LROC NAC and KPLO LUTI observations across diverse lunar regions and illumination conditions. LNEM preserves terrain contours and maintains stable reconstruction quality under varying viewing geometries.