Abstract
Neural networks can be effectively used for image processing tasks using 2D and 3D input data. When structured point clouds are processed, the training data can be annotated manually, or synthetically generated. For the localization or classification of objects bailed in bags and bundles, a parameterized and textured model of a bag or a physically-based bundling simulation can be used to generate synthetic training data. The generated dataset can be used for 6D pose estimation, classification of bags, or normal estimation of deformed bundles.