ComplementMe: Weakly-Supervised Component Suggestions for 3D Modeling

Assembly-based tools provide a powerful modeling paradigm for non-expert shape designers. However, choosing a component from a large shape repository and aligning it to a partial assembly can become a daunting task, especially, if the repository is not consistently segmented into semantically labeled parts. In this work we propose a new method for suggesting complementary components and their placement for an incomplete 3D part assembly. We train neural networks to address both the selection and placement challenges by first retrieving appropriate complements from low-dimensional embedding of all unlabeled components, and then choosing where to place each retrieved component. In both cases, due to the combinatorial nature of part arrangements, there is no single answer for the network outputs. Thus, our key technical insight is to predict a probability distribution over possible answers, which allows us to deal with ambiguities, enabling a smooth UI for integrating user preferences into the design process. We also demonstrate that our method can be used to design complex shapes with minimal or no user input. Finally, we propose a novel benchmark for evaluating component suggestion systems and show that our method outperforms state-of-the-art techniques.