SE(3)-Diffusion: An Equivariant Diffusion Model for 3D Point Cloud Generation

3D point cloud is a set of discrete data points in three-dimension space and it has been widely applied in computer vision, robotics, and autonomous driving, for tasks such as object recognition, tracking, and scene reconstruction, becoming an important data format for 3D perception and processing. On the other hand, deep generative model has achieve great success in image and text generation, however, due to its irregular structure and translational rotational equivariant properties, The generation of 3D point clouds still needs to be explored. Here we proposed a novel and effective model, which is able to extract translation-invariant and rotation-invariant feature of cloud points, learn the latent distribution of point cloud data, and then generate high-quality point cloud data. The experimental results prove that it can learn the real data distribution and generate high-quality point clouds.