Generative explainable models for 3D shapes

Collaborators: Hope Yao, Malcolm Regan, and Max Yi Ren

Abstract
In this ongoing project, we investigate model architectures and learning procedures that enable generation of novel 3D shapes based on unseen combinations of functional labels of objects. To do so, we look for robust classifiers enabled by iteratively learning through crowdsourcing mechanisms. In an existing work, we acquire 3D shape saliency knowledge using a crowdsourcing game, inspired by von Ahn's Peekaboom game.

Images

Related Papers
1. Yao, H., and Ren, Y. (2016). Impressionist: A 3D Peekaboo Game for Crowdsourcing Shape Saliency. In ASME 2016 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference. American Society of Mechanical Engineers. , Link