This project was awarded the Gold medal at EUROINVENT.
This project was awarded the Gold medal and Organizer's Choice Award at iCAN.
This project was a finalist at Genius Olympiad (Final rounds canceled).
3D-HisSite: 3D Reconstruction of Historic Sites Damaged by Drastic Climate Change via Image Inputs Utilizing Deep Learning
Recently, severe effects of drastic climate changes have eroded or destroyed historical sites, posing a grave risk in preserving them. Despite efforts to create 3D shapes of the remaining historic sites to bequeath these historical values digitally, current methods cannot restore the historic site’s former shape that does not currently exist or have already been damaged. In order to address these problems, we explore a complex deep learning network that aims to generate a 3D voxel model via image inputs. Inspired by Stanford University’s 3D-R2N2 network, our network can reconstruct the 3D shape of the historic site using only image inputs. The network is trained and validated via ShapeNet dataset and evaluated by the voxel IoU. The network is then fine-tuned with our Historical Site dataset and provides visualization through voxels. Our expectation is that future generations will be able to see and refer to our network in the work of restoration of destroyed or damaged historic sites.
View Project Paper