PS-Net:Point Shift Network For 3-D Point Cloud Completion

In recent times,point clouds have gained significant popularity as a means of describing 3D objects.However,the acquisition of point cloud data through scanning is often affected by equipment limitations and environmental factors,resulting in uneven density and missing parts of the point cloud shape.Consequently,the task of point cloud completion arises,aiming to infer the missing portions and reconstruct a complete shape from an incomplete point cloud.The availability of these complete point clouds can greatly enhance the performance of various downstream applications in remote sensing,including 3D reconstruction,3D printing,and autonomous driving,among others.The continuous development and advancement of deep learning has made it a great success in the field of computer vision,and the application of deep learning to point cloud data has become the mainstream way to achieve point cloud complementation.However,most existing methods based on deep learning cannot recover accurate structure details of the object.In this paper,we propose point shift network(PS-Net).Our main contributions lie in the following three-folds.First,we propose a multi-resolution point cloud encoder,which extracts and fuses multiresolution point cloud features hierarchically,thus avoiding information loss caused by a single global feature.Second,we design a multi-resolution point cloud generation structure,which can be combined with the multi-resolution point cloud decoder to generate gradually dense point clouds,avoiding the problem of non-uniformly density of the single-layer decoder.Third,we design the shift network,which is used to generate shift vectors to shift the coordinates of each point cloud,so as to further finetune the coordinate positions of point clouds,achieving more accurate prediction.This paper presents a comprehensive set of point cloud completion experiments conducted on the Shape Net,KITTI,and Scan Object NN datasets.The effectiveness of the proposed method is demonstrated through comparisons with state-of-the-art(SOTA)methods,both in terms of quantitative metrics and visualizations.Additionally,to validate the effectiveness of each component in the PS-Net,this paper includes quantitative evaluations,ablation experiments,and robustness tests for each component.The experimental results provide compelling evidence for the effectiveness of each component.Overall,this paper introduces a novel approach to point cloud completion,enabling accurate reconstruction of fine-grained point cloud shapes,and opens up new possibilities for research in autonomous driving,registration,and reconstruction domains.