We propose a novel method to synthesize geometric models from a given class of context-aware structured shapes such as buildings and other man-made objects. The central idea is to leverage powerful machine learning methods ...