Deformable 3D reconstruction from 2D images requires prior knowledge on the scene structure. Template-free methods use generic prior knowledge such as piecewise smoothness but require multiple image
