Point cloud completion aims to reconstruct detailed structures from partial observations. However, previous methods often suffered from inaccurate neighbourhood feature extraction and rough reconstruction of complex structures, which is difficult to complete detailed semantic shapes. To solve this problem, we present a method that applies dynamic transformers with adaptive neighbourhood feature fusion operations to resume complete point clouds. Firstly, we propose an adaptive neighbourhood feature extraction module, which contains a learnable global neighbourhood selection strategy and a traditional local k-nearest neighbour strategy to dynamically select neighbourhood points according to the shape of different objects. Secondly, we observed that traditional point generation methods based on folding-series operations limit their capacity of generating complex and faithful shapes. Inspired by cell division, we regard the process of reconstructing as point splitting and propose a genetic hierarchical point generation module, which means current points can inherit shape features from previous points and generate more detailed structures of their own. Extensive experiments of point cloud completion are carried out on PCN and Completion3D datasets to verify the effectiveness of our method. The average category chamfer distances of our method are 7.17(�10 −3 ) in PCN and 7.96(�10 −4 ) in Completion3D, which has better completion performance than other methods.
Completing point clouds from partial terrestrial laser scannings (TLS) is a fundamental step for many 3D visual applications, such as remote sensing, digital city and autonomous driving. However, existing methods mainly followed an ordinary auto-encoder architecture with only partial point clouds as inputs, and adopted K-Nearest Neighbors (KNN) operations to extract local geometric features, which takes insufficient advantage of input point clouds and has limited ability to extract features from long-range geometric relationships, respectively. In this paper, we propose a keypoints-aligned siamese (KASiam) network for the completion of partial TLS point clouds. The network follows a novel siamese auto-encoder architecture, to learn prior geometric information of complete shapes by aligning keypoints of complete-partial pairs during the stage of training. Moreover, we propose two essential blocks cross-attention perception (CAP) and self-attention augment (SAA), which replace KNN operations with attention mechanisms and are able to establish long-range geometric relationships among points by selecting neighborhoods adaptively at the global level. Experiments are conducted on widely used benchmarks and several TLS data, which demonstrate that our method outperforms other state-of-the-art methods by a 4.72% reduction of the average Chamfer Distance of categories in PCN dataset at least, and can generate finer shapes of point clouds on partial TLS data.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.