Diffusion Probabilistic Models for 3D Point Cloud Generation

TL;DR

This paper introduces a diffusion probabilistic model for 3D point cloud generation, inspired by thermodynamics, enabling effective shape synthesis, completion, and augmentation with competitive results.

Contribution

It proposes a novel diffusion-based approach for point cloud generation, modeling the reverse process as a conditioned Markov chain, with a closed-form variational bound for training.

Findings

Achieves competitive performance in point cloud generation

Demonstrates effective shape auto-encoding

Provides a new thermodynamics-inspired framework

Abstract

We present a probabilistic model for point cloud generation, which is fundamental for various 3D vision tasks such as shape completion, upsampling, synthesis and data augmentation. Inspired by the diffusion process in non-equilibrium thermodynamics, we view points in point clouds as particles in a thermodynamic system in contact with a heat bath, which diffuse from the original distribution to a noise distribution. Point cloud generation thus amounts to learning the reverse diffusion process that transforms the noise distribution to the distribution of a desired shape. Specifically, we propose to model the reverse diffusion process for point clouds as a Markov chain conditioned on certain shape latent. We derive the variational bound in closed form for training and provide implementations of the model. Experimental results demonstrate that our model achieves competitive performance in point cloud generation and auto-encoding. The code is available at \url{https://github.com/luost26/diffusion-point-cloud}.