A breakthrough in 3D preservation has come from a team of international researchers from Japan and China, who have developed a cutting-edge neural network model that can reconstruct 3D digital images of relief-type cultural heritage objects from old photographs. These reliefs, often found at historical sites across the globe, have suffered from a range of damage factors or simply deteriorated over time. Conventional restoration methods have proved to be extremely labor-intensive, as well as requiring specific niche knowledge, while this new technology provides a much more efficient solution.
Reliefs, which typically feature various figures protruding from a background, are usually admired from the front or side, making them suitable for 3D reconstruction from a single image. The model, created by a team led by Professor Satoshi Tanaka from Japan’s Ritsumeikan University and Dr. Jiao Pan from the University of Science and Technology Beijing, addresses challenges in depth estimation of the historical objects.
“We’ve enhanced depth estimation, particularly along soft edges, using a novel edge-detection approach,” explained Professor Tanaka, indicating their method’s improvements over previous attempts. The model successfully reconstructs reliefs with soaring accuracy, focusing on all the subtle variations in curvature that make up the so-called soft edges.
The scientists applied their model to reconstruct the “secret” sections of Borobudur Temple, a UNESCO World Heritage Site in Indonesia. Thanks to their model, the once-hidden reliefs can now be virtually explored through digital visualization and VR. This neural network definitely opens new horizons in cultural heritage preservation, as well as captivating, immersive experiences in the metaverse Lara Croft could envy.
