“3DShape2VecSet: A 3D Shape Representation for Neural Fields and Generative Diffusion Models” by Zhang, Tang, Nießner and Wonka
Conference:
Type:
Title:
- 3DShape2VecSet: A 3D Shape Representation for Neural Fields and Generative Diffusion Models
Session/Category Title: Diffusion for Geometry
Presenter(s)/Author(s):
Moderator(s):
Abstract:
This paper presents 3DShape2VecSet, a novel shape representation that encodes a 3D shape using neural fields as a set of vectors. The representation is designed for generative diffusion models and outperforms other methods in various 3D shape encoding and generative modeling tasks, including point-cloud completion and category/text/image-conditioned generation.
