Reconstruction of 3D Shapes from Cross-Sections using Indicator Networks

In this work, we propose to tackle the reconstruction of 3D objects from cross-sections with the help of deep learning, focusing on a data-driven approach to reconstruction. Typically, such techniques are used to learn high-level features of 2D images, where their applications range from classification, segmentation to point cloud reconstruction. They can be used in several 3D non-euclidean spaces, such as meshes, point clouds, but one there are no existing definitions for them in the case of cross-sections.