Toward editable implicit shapes
Abstract
Implicit shape representation is an old concept in computer graphics mainly used in the surface reconstruction stage. Recently, it has become extremely popular in the computer vision community partially thanks to its ability to represent arbitrary geometric topology and resolution. However, for editing, implicit representation is much less popular compared to its explicit counterpart, e.g., mesh or even points due to the lack of locality, interpretability, and controllability. Although one can convert the implicit representation to an explicit representation for editing, directly editing in the implicit domain promises a seamlessly integration with many exciting reconstruction pipelines in computer vision. In this talk, I’ll present two of my recent works that attempt to ease of shape editing with implicit representations, as well as discuss some of the biggest challenges in this domain.
