Deep Unsupervised Learning of 3D Point Clouds via Graph Topology Inference and Filtering

Abstract

We propose a deep autoencoder with graph topology inference and filtering to achieve compact representations of unorganized 3D point clouds in an unsupervised manner. Many previous works discretize 3D points to voxels and then use latticebased methods to process and learn 3D spatial information; however, this leads to inevitable discretization errors. In this work, we try to handle raw 3D points without such compromise. The encoder of the proposed networks adopts similar architectures as in PointNet, which is a well-acknowledged method for supervised learning of 3D point clouds. The decoder of the proposed networks involves three novel modules: the folding module, the graph-topology-inference module, and the graphfiltering module. The folding module folds a canonical 2D lattice to the underlying surface of a 3D point cloud, achieving coarse reconstruction; the graph-topology-inference module learns a graph topology to represent pairwise relationships between 3D points, pushing the latent code to preserve both coordinates and pairwise relationships of points in 3D point clouds; and the graph-filtering module designs graph filters based on the learnt graph topology and refines the coarse reconstruction to obtain the final reconstruction. We further provide theoretical analyses of the proposed architecture. We provide an upper bound for the reconstruction loss and further show the superiority of graph smoothness over spatial smoothness as a prior to model 3D point clouds. In the experiments, we validate the proposed networks in three tasks, including 3D point clouds reconstruction, visualization, and transfer classification. The experimental results show that (1) the proposed networks outperform the state-ofthe-art methods in various tasks, including reconstruction and transfer classification; (2) a graph topology can be inferred as auxiliary information without specific supervision on graph topology inference; and (3) graph filtering refines the reconstruction, leading to better performances.

3D Grid Transformation Network For Point Cloud Completion

“…The folding method uses MLPs to transform the 2D grid points to the surface of the point clouds. However, this results in a default and uniform plane pattern for the generated 3D points [17]. So, we need Fig.…”

MM-Flow

“…MLP based completion networks: Pioneered by PointNet [24], many approaches are introduced to work directly on point cloud to avoid geometric information loss [3,36,43,44]. These methods apply the shared multi-layer perceptron (MLP) to extract features and the max-pooling layer to solve the permutation invariant problem.…”

Soft Delivery: Survey on A New Paradigm for Wireless and Mobile Multimedia Streaming

“…Fig. 12 showed their GNN-based auto-encoder (GAE) [192], [193] to encode 3D point clouds into a limited number of latent variables. One of the benefits of the GAE is that it allows graph signal reconstruction from a limited number of latent variables without requiring additional metadata.…”