“Neural Wavelet-domain Diffusion for 3D Shape Generation” by Hui, Li, Hu and Fu
Conference:
Type(s):
Title:
- Neural Wavelet-domain Diffusion for 3D Shape Generation
Session/Category Title:
- Shape Generation
Presenter(s)/Author(s):
Abstract:
This paper presents a new approach for 3D shape generation, enabling a direct generative modeling on a continuous implicit representation in wavelet frequency domain. Specifically, we propose a compact wavelet representation with a pair of coarse and detail coefficient volumes to implicitly represent 3D shapes via truncated signed distance function and multi-scale biorthogonal wavelet. Then, we formulate a pair of neural networks: a generator based on the diffusion model for producing diverse shapes in the form of coarse coefficient volume; and a detail predictor to further produce compatible detail coefficient volumes for enriching the generated shapes with fine details. Both quantitative and qualitative experimental results manifest the superiority of our approach in generating diverse and high-quality shapes with complex topology and structures, clean surfaces, and fine details, exceeding the 3D generation capabilities of the state-of-the-art models.
