中图分类号：TP391.4 文献标志码：A DOI: 10.16157/j.issn.0258-7998.257077 中文引用格式： 林志浩，赵家池，程卓. 基于深度生成模型的点云生成算法综述[J]. 电子技术应用，2026，52(2)：7-14. 英文引用格式： Lin Zhihao，Zhao Jiachi，Cheng Zhuo. A survey of point cloud generation algorithms based on deep generative models[J]. Application of Electronic Technique，2026，52(2)：7-14.
A survey of point cloud generation algorithms based on deep generative models
Lin Zhihao1，2，Zhao Jiachi1，2，Cheng Zhuo1，2
1.College of Information Science and Engineering， Ningbo University；2.Key Laboratory of Mobile Network Application Technology of Zhejiang Province
Abstract： As a core task in the field of 3D vision, point cloud generation plays a vital role in scenarios such as point cloud shape completion, upsampling, and synthesis. It is widely used in key application areas including autonomous driving, robot navigation, and medical imaging. Due to the inherent unordered nature, sparsity, and structural complexity of point cloud data, traditional geometric modeling methods struggle to efficiently generate high-quality and diverse point cloud samples. In recent years, point cloud generation techniques based on deep generative models have developed rapidly and become a research hotspot, significantly improving the quality and efficiency of point cloud generation. This paper reviews the latest progress and current challenges in point cloud generation algorithms based on deep generative models and discusses potential future research directions.
Key words : point cloud generation；deep generative model；generative adversarial network （GAN）；variational autoencoder （VAE）；normalizing flow；autoregressive model；diffusion model
