Geometric datasets span virtual environments, characters, design graphics, medical, and scientific models. These data can fuel powerful tools for scene understanding, creation of expressive visual artifacts, and analysis of complex phenomenon. Rapid development of novel neural network architectures and representations provides essential building blocks that can be re-assembled and re-used for a wide range of applications. The cost of this versatility is that significant efforts go into choosing the right blocks for a particular task. In this talk I will describe shape representations that can be used for generative and analysis tasks, and will propose a neural network architecture that efficiently unifies various representations for diverse tasks.
Back to Workshop II: Shape Analysis