Patch-based Methods in Data-Driven Geometry Processing
Date:
The recent advancements of deep learning led to powerful models for image and text processing. On the other hand, methods for 3D geometric data failed to keep track with their Euclidean counterparts. In this talk, we will discuss how patches, i.e. local regions of 3D surfaces, may be employed to develop general geometric neural models. We illustrate this perspective through recent algorithms whose core principles rely on local geometric structure, highlighting how patch-based representations enable expressive and scalable procedures and learning on 3D shapes.