Deep neural networks have achieved great progress in 3D scene understanding. However, recent methods mainly focused on objects with canonical orientations contrast random postures reality. In this letter, we propose a hierarchical network, named Local Frame Network (LFNet), based the local rotation invariant coordinate frame for robust point cloud analysis. The patches different orientated are transformed into an identical distribution frame, and coordinates taken as input features to...

Autoscanning of an unknown environment is the key to many AR/VR and robotic applications. However, autonomous reconstruction with both high efficiency quality remains a challenging problem. In this work, we propose reconstruction-oriented autoscanning approach, called ScanBot, which utilizes hierarchical deep reinforcement learning techniques for global region-of-interest (ROI) planning improve scanning local next-best-view (NBV) enhance quality. Given partially reconstructed scene, policy...

Recently, deep neural networks have made remarkable achievements in 3D point cloud analysis. However, the current shape descriptors are inadequate for capturing information thoroughly. To handle this problem, a feature representation learning method, named Dual-Neighborhood Deep Fusion Network (DNDFN), is proposed to serve as an improved encoder task of Specifically, traditional local neighborhood ignores long-distance dependency and DNDFN utilizes adaptive key replenishment mechanism...

Parkinson's disease (PD) is one of the common progressive neurodegenerative disorder with motor deficits. A substantial volume research has reported various methods to characterize rat model, but performances are unsatisfactory. Because complexity behavior data, we can hardly recognize non-trivial behavioral traits and distinguish difference between ill healthy.

Traditional Convolutional Neural Networks (CNN) are limited to extract informative local features of point clouds due the fixed geometric structures in convolution kernel against irregular and unstructured clouds. It usually requires data transformation such as voxelization or projection, inducing a possible loss information. Instead fitting input points by regularization, we choose fit conduct convolution. In this paper, present new method define compute directly on 3D Adaptive Surface...

The tasks of point cloud analysis are very challenging. Designing efficient convolution operation is the key to accomplish these tasks. In order capture structure information, neighborhood usually needs be considered when designing convolution. At present, most works adopt K-Nearest Neighbor or ball query construct neighborhood. However, two methods only focus on spatial distance relationship and ignore long-distance dependence between points. this paper, Learnable-Graph Convolutional Neural...

Transformer plays an increasingly important role in various computer vision areas and remarkable achievements have also been made point cloud analysis. Since they mainly focus on point-wise transformer, adaptive channel encoding transformer is proposed this paper. Specifically, a convolution called Transformer-Conv designed to encode the channel. It can feature channels by capturing potential relationship between coordinates features. Compared with simply assigning attention weight each...

Recently, deep neural networks have made remarkable achievements in 3D point cloud classification. However, existing classification methods are mainly implemented on idealized clouds and suffer heavy degradation of per-formance non-idealized scenarios. To handle this prob-lem, a feature representation learning method, named Dual-Neighborhood Deep Fusion Network (DNDFN), is proposed to serve as an improved encoder for the task DNDFN utilizes trainable neighborhood method called TN-Learning...

Attention mechanism plays a more and important role in point cloud analysis channel attention is one of the hotspots. With so much information, it difficult for neural networks to screen useful information. Thus, an adaptive encoding proposed capture relationships this paper. It improves quality representation generated by network explicitly interdependence between channels its features. Specifically, channel-wise convolution (Channel-Conv) adaptively learn relationship coordinates features,...