3D reconstruction of Borobudur reliefs from 2D monocular photographs based on soft-edge enhanced deep learning

A groundbreaking deep learning method achieves 97% accuracy in reconstructing hidden Borobudur temple reliefs from old monocular photographs, addressing the challenge of digitizing cultural heritage sites inaccessible to traditional 3D scanning. Unlike existing models optimized for sharp-edged scenes, this approach leverages soft-edge features to estimate depth with precision, even in low-curvature areas like the buried reliefs behind stone walls. By integrating reconstructed data with visible 3D scans, researchers enable 'see-through' visualizations that simultaneously display both visible and previously hidden regions of the UNESCO World Heritage Site.

Pan, J., Li, L., Yamaguchi, H., Hasegawa, K., Thufail, F. I., Brahmantara, & Tanaka, S. (2022). 3D reconstruction of Borobudur reliefs from 2D monocular photographs based on soft-edge enhanced deep learning. ISPRS Journal of Photogrammetry and Remote Sensing, 183, 439–450. https://doi.org/10.1016/j.isprsjprs.2021.11.007