PointGrow: Autoregressively Learned Point Cloud Generation with Self-Attention

TL;DR

PointGrow is an autoregressive model that uses self-attention to generate diverse, realistic 3D point clouds, capturing long-range dependencies and enabling applications like shape manipulation.

Contribution

The paper introduces PointGrow, a novel autoregressive point cloud generator that incorporates self-attention to better model inter-point relations and long-range dependencies.

Findings

Achieves high realism and diversity in point cloud generation

Effective in both unconditional and conditional tasks

Enables applications like unsupervised feature learning and shape arithmetic

Abstract

Generating 3D point clouds is challenging yet highly desired. This work presents a novel autoregressive model, PointGrow, which can generate diverse and realistic point cloud samples from scratch or conditioned on semantic contexts. This model operates recurrently, with each point sampled according to a conditional distribution given its previously-generated points, allowing inter-point correlations to be well-exploited and 3D shape generative processes to be better interpreted. Since point cloud object shapes are typically encoded by long-range dependencies, we augment our model with dedicated self-attention modules to capture such relations. Extensive evaluations show that PointGrow achieves satisfying performance on both unconditional and conditional point cloud generation tasks, with respect to realism and diversity. Several important applications, such as unsupervised feature learning and shape arithmetic operations, are also demonstrated.