The C3D company produces teeth models for optimal treatment for broken teeth. The teeth models were built using a 3D imaging by stereo images pairs. Generally, to conduct 3D reconstruction out of stereo pairs, points matching must be first applied. Teeth images characterized with very low texture and therefor low on features that may aid with the matching task. In order to overcome this obstacle, C3D uses a micro-powder producing artificial texture on the teeth. C3D wants to dismiss the usage of this powder and overcome this issue using state of the art deep learning methods.
In this project, we tested the ability of producing a 3D model out of stereo pairs using deep-learning based method. These algorithms were developed by researches to solve a generic 3D reconstruction problem and were tested with common datasets (e.g. KITTI). These algorithms were modified and changed to be compatible with the teeth dataset (special optics).