Farthest Point Research Articles

Tree Species Classification Using Ground-Based LiDAR Data by Various Point Cloud Deep Learning Methods

Tree species information is an important factor in forest resource surveys, and light detection and ranging (LiDAR), as a new technical tool for forest resource surveys, can quickly obtain the 3D structural information of trees. In particular, the rapid and accurate classification and identification of tree species information from individual tree point clouds using deep learning methods is a new development direction for LiDAR technology in forest applications. In this study, mobile laser scanning (MLS) data collected in the field are first pre-processed to extract individual tree point clouds. Two downsampling methods, non-uniform grid and farthest point sampling, are combined to process the point cloud data, and the obtained sample data are more conducive to the deep learning model for extracting classification features. Finally, four different types of point cloud deep learning models, including pointwise multi-layer perceptron (MLP) (PointNet, PointNet++, PointMLP), convolution-based (PointConv), graph-based (DGCNN), and attention-based (PCT) models, are used to classify and identify the individual tree point clouds of eight tree species. The results show that the classification accuracy of all models (except for PointNet) exceeded 0.90, where the PointConv model achieved the highest classification accuracy for tree species classification. The streamlined PointMLP model can still achieve high classification accuracy, while the PCT model did not achieve good accuracy in the tree species classification experiment, likely due to the small sample size. We compare the training process and final classification accuracy of the different types of point cloud deep learning models in tree species classification experiments, further demonstrating the advantages of deep learning techniques in tree species recognition and providing experimental reference for related research and technological development.

Experimental study, dimensional analysis and an integral model for horizontal buoyant turbulent jet fires under opposing wind

This work demonstrates a comprehensive model for horizontal turbulent jet flame geometries under opposing wind, for which few data or models can be found in the literature. This situation can be a practical scenario of offshore drilling platform with a horizontal flare under opposing wind. The opposing wind pushes the jet flame back, which eventually turns around and upwards, pointing finally downstream in the windward direction. The flame geometries are characterized by four parameters including horizontal length and vertical height from nozzle exit to the location of farthest point that flame envelope could reach before the flame turns around as well as horizontal length and vertical height from nozzle exit to the location of flame tip. Experiments are carried out for horizontal jets discharged from circular nozzles having diameters of 3, 5, 7 and 9 mm and employing propane as fuel with opposing wind speeds from 0.35 to 1.08 m/s. The locations of flame tip and the flame farthest point decrease with increasing opposing wind for a constant heat release rate. Dimensional analysis is derived from a physical model considering the momentum of fuel jet, flame buoyancy due to the flame temperature rise, opposing wind momentum and normalized flow rate at stoichiometric conditions at flame tip. The normalized flame geometry parameters based on the proposed model are well represented by two characteristic length scales derived by accounting for the interaction between excess jet-wind momentum and flame buoyancy as well as that between opposing wind momentum and flame buoyancy. In addition, the normalized mass flow rate at stoichiometric conditions determines the total length of flame trajectory. Moreover, an integral model is deduced by an air entrainment model and top hat profiles for the mass and momentum conservations to predict well the flame geometries of horizontal jet fires under opposing wind.

Characterization of Microplastics in Total Atmospheric Deposition Sampling from Areas Surrounding Industrial Complexes in Northwestern Colombia

In this study, the total atmospheric deposition of microplastics in places near an industrial complex in Cartagena, Colombia is evaluated by taking samples at three points at different distances from the industrial zone. The samples obtained were treated to quantify and identify the type of plastic present, indicating the presence of polystyrene, polypropylene, polyethylene, polyvinyl chloride, and polyethylene terephthalate. The values of microplastics obtained were related to the climatic conditions of the areas (wind and rainfall) to determine these effects on the transport of microplastics. Values of 9472 ± 702 MP were obtained at the point closest to the industrial area and 1455 ± 14 MP at the farthest point, within which a greater proportion were fibers. In this study, it was also possible to determine that the microplastics most affected by the wind were the fragments.

NGD-Transformer: Navigation Geodesic Distance Positional Encoding with Self-Attention Pooling for Graph Transformer on 3D Triangle Mesh

Following the significant success of the transformer in NLP and computer vision, this paper attempts to extend it to 3D triangle mesh. The aim is to determine the shape’s global representation using the transformer and capture the inherent manifold information. To this end, this paper proposes a novel learning framework named Navigation Geodesic Distance Transformer (NGD-Transformer) for 3D mesh. Specifically, this approach combined farthest point sampling with the Voronoi segmentation algorithm to spawn uniform and non-overlapping manifold patches. However, the vertex number of these patches was inconsistent. Therefore, self-attention graph pooling is employed for sorting the vertices on each patch and screening out the most representative nodes, which were then reorganized according to their scores to generate tokens and their raw feature embeddings. To better exploit the manifold properties of the mesh, this paper further proposed a novel positional encoding called navigation geodesic distance positional encoding (NGD-PE), which encodes the geodesic distance between vertices relatively and spatial symmetrically. Subsequently, the raw feature embeddings and positional encodings were summed as input embeddings fed to the graph transformer encoder to determine the global representation of the shape. Experiments on several datasets were conducted, and the experimental results show the excellent performance of our proposed method.

Effect of Medical Image Fusion in the Treatment of Poststroke Limb Dysfunction with Acupuncture and Moxibustion of Traditional Chinese Medicine.

According to relevant data, the morbidity and mortality of strokes in China remain high. Without effective treatment, stroke morbidity and mortality will continue to rise, which may become the second leading disease in the world. With the nonstop advancement and improvement of clinical innovation in China, the death pace of stroke patients has dropped altogether. After clinical treatment, the patient actually showed a progression of sequelae, which made it challenging to work on the personal satisfaction of the patient. The purpose for this paper was to concentrate on the impact of medical image fusion in the treatment of poststroke appendage brokenness with TCM needle therapy. The related concepts of medical image fusion and the meaning of acupuncture and moxibustion in traditional Chinese medicine, stroke, and limb dysfunction were introduced. In this study, acupuncture and moxibustion were analyzed to explore the therapeutic effect of this type of therapy on upper extremity dysfunction caused by phlegm and blood stasis blocking collaterals and to provide a scientific method for the treatment and efficacy judgment of upper extremity motor dysfunction after stroke. Before the treatment measures were taken, there was no significant difference in the general data and all index scores between the two groups (P > 0.05), and there was no significant difference in the baseline data, reflecting high balance and comparability. In the following 3 months of treatment, the FMA score, NIHSS score, BI list, and VAS score of the two groups of patients were essentially not quite the same as those before treatment (P < 0.05). When treatment, there was a huge contrast between the trial group and the control group (P < 0.05). The finish of the trial in this paper is that needle therapy joined with pricking and measuring can essentially work on the engine capability of stroke patients with furthest point brokenness brought about by mucus and blood balance impeding securities.

LoRa Network-Based System for Monitoring the Agricultural Sector in Andean Areas: Case Study Ecuador.

This article focuses on the development of a system based on the long-range network (LoRa), which is used for monitoring the agricultural sector and is implemented in areas of the Andean region of Ecuador. The LoRa network is applied for the analysis of climatic parameters by monitoring temperature, relative humidity, soil moisture and ultraviolet radiation. It consists of two transmitter nodes and one receiver node, a LoRa Gateway with two communication channels for data reception and one for data transmission, and an IoT server. In addition, a graphical user interface has been developed in Thinger.io to monitor the crops and remotely control the actuators. The research conducted contains useful information for the deployment of a LoRa network in agricultural crops located in mountainous areas above 2910 m.a.s.l., where there are terrains with irregular orography, reaching a coverage of 50 hectares and a range distance of 875 m to the farthest point in the community of Chirinche Bajo, Ecuador. An average RSSI of the radio link of −122 dBm was obtained in areas with a 15% slope and 130 m difference in height according to the Gateway, where the presence of vegetation, eucalyptus trees and no line-of-sight generated interference to the radio signal. The success rate of PDR packet delivery with an SF of nine, had a better performance, with values of no less than 76% and 92% in uplink and downlink respectively. Finally, the technological gap is reduced, since the network reaches places where traditional technologies do not exist, allowing farmers to make timely decisions in the production process in the face of adverse weather events.

SSA3D: Semantic Segmentation Assisted One-Stage Three-Dimensional Vehicle Object Detection

One-stage 3D object detection using mobile light detection and ranging (LiDAR) has developed rapidly in recent years. Specifically, one-stage methods have attracted attention because of their high efficiency and light weight compared with two-stage methods. Inspired by this, we present the semantic segmentation assisted one-stage three-dimensional vehicle object detection (SSA3D), a network for the rapid detection of objects that keeps the advantages of the semantic segmentation module in the two-stage methods without increasing redundant computational load. First, we modified the sampling of the farthest point to improve the quality of the sampling points. This helps to reduce sampling outlier points and bad points that are difficult to perceive in the spatial structure information surrounding the point. Second, a neighbor attention group module is devoted to selectively add extra weight to neighbor points because of the different importance of neighbor points for the corresponding sampling point. Correctly increasing the weight is helpful to obtain richer spatial structure information. Finally, a delicate box generation module is included as a voted center point layer based on the generalized Hoff vote method and an anchor-free regression. We used the feature aggregation module as the backbone and the feature propagation module as the auxiliary network to achieve efficiency. At the same time, the auxiliary network retains the ability to extract point-wise features from the state-of-the-art semantic segmentation network. In experiments, we evaluated and tested the SSA3D on a common KITTI dataset and achieved improved performance in the class of car accuracy.

Morphometric Analysis and Clinical Significance of Cervical Vertebrae: Example of Turkey

Objectives: The spine is the column that carries the weight of the head and the torso and contains the medulla spinalis that is a part of the Central Nervous System within the canal inside it. Changes occur in the anatomical structures of the vertebrae in cases of infections involving the vertebrae and fractures and deformities arising from traumatic or non-traumatic causes. The determination of such changes in the vertebrae is critically important in terms of treatment or surgical intervention. Morphometric measurements have an important place in the detection of these changes. Moreover, recently, cervical vertebral measurements have been used in sex identification, the preliminary diagnosis of genetic diseases and age identification. We aimed for the results of our study to support clinical interventions to be made in the cervical vertebrae, forensic medicine applications and anthropological applications as a reference in the literature. Methods: In the study, 54 cervical vertebrae in the form of dry bones belonging to the neck region were used as the material. Twenty-three different parameters were measured with a digital caliper at a precision 0.01 mm. Results: Measurements were made on 31 C3-C6, 7 C7, 8 C1 and 8 C2 vertebrae. The mean measurement values were determined as a corpus vertebrae height (anterior) of 17.26±2.10 mm, a corpus vertebrae length (posterior) of 14.81±2.02 mm, a right lamina arcus vertebrae length of 13.34±2.11 mm, a diagonal foramen vertebrale length of 20.21±1.60 mm, a distance between the farthest points of massa lateralis atlantis of 73.99 (66.86-86.94) mm, and a transverse corpus vertebrae diameter of 25.03±4.35 mm. Conclusion: In the cervical vertebral measurements of the Turkish population in our study, we observed that the results on corpus vertebrae height and transverse diameter varied based on races, and the measurements of the Turkish population were higher. In addition, low, medium and high positive-negative relationships were determined by performing correlation analysis between the vertebrae. Accordingly, we think that these analyses will be helpful in the preparation of the atlas and the drawing of vertebrae.

End-to-End Point Cloud Completion Network with Attention Mechanism.

We propose a conceptually simple, general framework and end-to-end approach to point cloud completion, entitled PCA-Net. This approach differs from the existing methods in that it does not require a “simple” network, such as multilayer perceptrons (MLPs), to generate a coarse point cloud and then a “complex” network, such as auto-encoders or transformers, to enhance local details. It can directly learn the mapping between missing and complete points, ensuring that the structure of the input missing point cloud remains unchanged while accurately predicting the complete points. This approach follows the minimalist design of U-Net. In the encoder, we encode the point clouds into point cloud blocks by iterative farthest point sampling (IFPS) and k-nearest neighbors and then extract the depth interaction features between the missing point cloud blocks by the attention mechanism. In the decoder, we introduce a new trilinear interpolation method to recover point cloud details, with the help of the coordinate space and feature space of low-resolution point clouds, and missing point cloud information. This paper also proposes a method to generate multi-view missing point cloud data using a 3D point cloud hidden point removal algorithm, so that each 3D point cloud model generates a missing point cloud through eight uniformly distributed camera poses. Experiments validate the effectiveness and superiority of PCA-Net in several challenging point cloud completion tasks, and PCA-Net also shows great versatility and robustness in real-world missing point cloud completion.

TATA GUNA LAHAN DAN KEJADIAN BANJIR DI KAWASAN PERKOTAAN KABUPATEN TABALONG

The Regional Spatial Plan (RTRW) is a direction in space utilization and control of space utilization, but the RTRW still uses a small scale of 1: 50,000, which means that information on topographical conditions is not described in detail, so the housing planning and licensing are not informed in detail as a result, low-lying areas Initially functioned as a temporary water reservoir turned into a residential area, it was unavoidable. The conventional drainage concept currently applied in Tanjung urban area is an effort to dispose of or drain excess water as soon as possible to the nearest river, namely the Mangkusip River. Along with the development which is not balanced with good drainage planning causes water to flow improperly, and high-intensity rainfall may occur flooding. The purpose of the research is to examine land use in urban areas with three-dimensional spatial data related to flood events, analyze flood events in urban areas, and determine locations as protected areas for water conservation. Three-dimensional spatial data using DEM from DEMNAS, the digital image in 2020. Determining protected location areas (jungle cities) using several parameters, where each parameter has its score value, both analyzes utilize GIS. Counting of a flood using Rational and Manning method. The results showed: 1) Topographic conditions and the slope of the research site, 2) Natural formation network of the Mangkusip river and tributary, 3) Length (L) Length of air traffic from the farthest point to the point at the review point, 4) The catchment area (A) from the modeling results, 5) The average slope of the water path from the farthest point to the observation point. 6) Land use (C), 7) The runoff flood discharge is greater than the flood discharge from the Mangkusip river reservoir, resulting in flooding, 8) From the results of the assessment with overlay scoring, the result is an area of 59,318 hectares for water conservation areas to prevent flooding.

Interpreting and Georeferencing the Concept of “Near” in Locality Descriptors in Biodiversity Records

The process of georeferencing is fundamentally a matter of spatial relationships: where is the point of interest in relation to the Prime Meridian, the Equator, the nearest landmark, etc. The process of adding geographic coordinates and uncertainty measurements to data coming from the work of others is often complicated by interpretation and subjectivity. Take for instance the description, “found near the city of Springfield", which might be assigned a coordinate pair based on the centroid of the city’s footprint. If the implication is that the subject was found around Springfield, then how close? Could this description refer to a point within the city? Is this description referring to the city limits, the farthest reaches of city mailing addresses, or the original author’s interpretation of the territory of Springfield? At present, best practices for adding coordinate uncertainty to such data records is to expand arbitrarily the boundaries of the feature (e.g., the city limits of Springfield), and use the distance from this new shape’s center to its farthest point (Zermoglio et al. 2020). In this talk, I propose a way to objectively quantify the area that can be considered “near” a given reference point. Voronoi diagrams are collections of points (or “cells”) that correspond to a single region. Any coordinate location within a given region is closer to that region’s cell than to any other cell in the diagram (Dobrin 2005). I describe possible applications of Voronoi diagrams in georeferencing sites described as “near” named places, which take into account the uneven distribution of named places across a region. I demonstrate the straightforward process of generating these diagrams using readily available open-source GIS (Geographic Information System) data and programs, and discuss both shortcomings and possible deeper-level applications of these approaches.

Elevated soil and blood lead levels with increasing residential proximity to a mine tailings facility in Soweto, South Africa

Mining has long been associated with the release of a range of toxic metals including lead, elevated levels of which have been found in land surrounding mine dumps as far afield as Spain, India, Macedonia and Nigeria. Mining has been a key pillar of economic development in South Africa for around a century and a half, and has left a legacy of major environmental contamination, with the poorest experiencing the highest burden of exposure. The current study was undertaken to ascertain the concentrations of lead in soil and in the blood of adults and children residing at increasing distances from a cluster of large tailings facilities (MTFs) in greater Johannesburg, South Africa. Through a cross-sectional survey data on household characteristics were collected through a questionnaire survey, and supplemented with samples of soil from household gardens, together with blood samples from child-adult pairs from all households included in the study. Soil lead concentrations decreased significantly (median = 35.7 mg/kg closest to the mine tailings facility (MTF) and 8.6 mg/kg at the furthest point in the study site) with increasing distance from the MTF. Blood lead levels were highest in both adults (median = 3.0 μg/dL) and children (median = 4.0 μg/dL) who lived closest (≤500 m) to the MTF, and lowest (1.4 μg/dL in adults and 2.5 μg/dL in children) in those who lived furthest away (4–5 km). The study findings point to a need for greater emphasis on the precautionary principle in environmental health and for health impact assessments to inform decisions on planning, especially with regard to the location of human settlements relative to major, polluting development initiatives.

DGPolarNet: Dynamic Graph Convolution Network for LiDAR Point Cloud Semantic Segmentation on Polar BEV

Semantic segmentation in LiDAR point clouds has become an important research topic for autonomous driving systems. This paper proposes a dynamic graph convolution neural network for LiDAR point cloud semantic segmentation using a polar bird’s-eye view, referred to as DGPolarNet. LiDAR point clouds are converted to polar coordinates, which are rasterized into regular grids. The points mapped onto each grid distribute evenly to solve the problem of the sparse distribution and uneven density of LiDAR point clouds. In DGPolarNet, a dynamic feature extraction module is designed to generate edge features of perceptual points of interest sampled by the farthest point sampling and K-nearest neighbor methods. By embedding edge features with the original point cloud, local features are obtained and input into PointNet to quantize the points and predict semantic segmentation results. The system was tested on the Semantic KITTI dataset, and the segmentation accuracy reached 56.5%

Tree Species Classification of Backpack Laser Scanning Data Using the PointNet++ Point Cloud Deep Learning Method

To investigate forest resources, it is necessary to identify the tree species. However, it is a challenge to identify tree species using 3D point clouds of trees collected by light detection and ranging (LiDAR). PointNet++, a point cloud deep learning network, can effectively classify 3D objects. It is important to establish high-quality individual tree point cloud datasets when applying PointNet++ to identifying tree species. However, there are different data processing methods to produce sample datasets, and the processes are tedious. In this study, we suggest how to select the appropriate method by designing comparative experiments. We used the backpack laser scanning (BLS) system to collect point cloud data for a total of eight tree species in three regions. We explored the effect of tree height on the classification accuracy of tree species by using different point cloud normalization methods and analyzed the effect of leaf point clouds on classification accuracy by separating the leaves and wood of individual tree point clouds. Five downsampling methods were used: farthest point sampling (FPS), K-means, random, grid average sampling, and nonuniform grid sampling (NGS). Data with different sampling points were designed for the experiments. The results show that the tree height feature is unimportant when using point cloud deep learning methods for tree species classification. For data collected in a single season, the leaf point cloud has little effect on the classification accuracy. The two suitable point cloud downsampling methods we screened were FPS and NGS, and the deep learning network could provide the most accurate tree species classification when the number of individual tree point clouds was in the range of 2048–5120. Our study further illustrates that point-based end-to-end deep learning methods can be used to classify tree species and identify individual tree point clouds. Combined with the low-cost and high-efficiency BLS system, it can effectively improve the efficiency of forest resource surveys.

AGS-SSD: Attention-Guided Sampling for 3D Single-Stage Detector

3D object detection based on LiDAR point cloud has always been challenging. Existing point cloud downsampling approaches often use heuristic algorithms such as farthest point sampling (FPS) to extract the features from a massive raw point cloud. However, FPS has disadvantages such as low operating efficiency and inability to sample key areas. This paper presents an attention-guided downsampling method for point-cloud-based 3D object detection, named AGS-SSD. The method contains two modules: PEA (point external attention) and A-FPS (attention-guided FPS). PEA explores the correlation between the data and uses the external attention mechanism to extract the semantic features in the set abstraction stage. The semantic information, including the relationship between the samples, is sent to the candidate point generation module as context points. A-FPS weighs the point cloud according to the generated attention map and samples the foreground points with rich semantic information as candidate points. The experimental results show that our method achieves significant improvements with novel architectures against the baseline and runs at 24 frames per second for inference.

DeepMatch: Toward Lightweight in Point Cloud Registration.

From source to target, point cloud registration solves for a rigid body transformation that aligns the two point clouds. IterativeClosest Point (ICP) and other traditional algorithms require a long registration time and are prone to fall into local optima. Learning-based algorithms such as Deep ClosestPoint (DCP) perform better than those traditional algorithms and escape from local optimality. However, they are still not perfectly robust and rely on the complex model design due to the extracted local features are susceptible to noise. In this study, we propose a lightweight point cloud registration algorithm, DeepMatch. DeepMatch extracts a point feature for each point, which is a spatial structure composed of each point itself, the center point of the point cloud, and the farthest point of each point. Because of the superiority of this per-point feature, the computing resources and time required by DeepMatch to complete the training are less than one-tenth of other learning-based algorithms with similar performance. In addition, experiments show that our algorithm achieves state-of-the-art (SOTA) performance on both clean, with Gaussian noise and unseen category datasets. Among them, on the unseen categories, compared to the previous best learning-based point cloud registration algorithms, the registration error of DeepMatch is reduced by two orders of magnitude, achieving the same performance as on the categories seen in training, which proves DeepMatch is generalizable in point cloud registration tasks. Finally, only our DeepMatch completes 100% recall on all three test sets.

Rejection of Plaint in the Light of Precedents of Superior Courts of Pakistan

This article delineates the dismissal of the plaint under Rule 11 Order VII, Civil Code Procedure (1908) as applied by the Superior Courts of Pakistan. The different general standards, for example, the inclination of the rightness of plaint. The valuation of the suits for court charge and locale,the obligation of the Court staff qua valuation, and prudence of the court to expand fixed time combined with various auxiliary viewpoints have likewise been examined. The most significant and summoned ground for dismissal of plaint-'banned by law' has been explained with representations zeroing in on the undertone of law and recognizing dismissal of the plaint and excusal of the suit. From there on, the outcomes of the dismissal of plaint and cures there against have additionally been brought getting looked at. The furthest point of this article closes the entire conversation.

Computational fluid dynamics study of the rear geometry influence on aerodynamic load coefficients of heliostats in stow position

The study of heliostat load coefficients in stow position has become more extensive because of the importance of this scenario in the design process and cost estimation of an entire Concentrating Solar Thermal Power Tower plant. When in stow position, in the presence of high-velocity wind flow, heliostats behavior can be compared to an airfoil; as the air flows at different velocities over the upper and lower sections, significant pressure imbalances need to be considered. This paper presents a Large Eddy Simulation study of how four different geometries in the rear area of a squared heliostat at a horizontal inclination affect the magnitudes of the hinge moment (CMHy), drag (CFx), lift (CFz) and overturning (CMy) coefficients. Results show maximum differences of more than 128% for peak CFx when comparing type A and B geometries at the same 45° wind orientations. Also, for types A and B at β = 180°, peak CMy and CMHy presented differences of 47% and 36%, respectively, between heliostat models. Also, considering force's change in direction (i.e. lift vs downforce), a 219% maximum difference was observed for peak CFz between type A and type D models at β = 45°. The results show how the differences in shapes and size proportions between the front and rear heliostat sections affect the stow position's aerodynamic load coefficients (ALC). It was also observed that the ratio of the chord length to the maximum vertical distance from the rear mirror surface to the farthest point of structure geometry of the heliostat is directly correlated with ALC variations for analyzed heliostats.

CLUSTERIZATION METHOD BASED ON BREADTH FIRST SEARCH OR BFS FOR A GRAPH

Clusterization is one of the types of algorithms of unsupervised learning. The idea behind it is that an algorithm learns patterns from untagged data. Such type of algorithm helps to find unseen dependencies in the untagged data itself. This paper presented algorithms based on Breadth-First Search or BFS for a Graph. The method was built based on the basic theory of clusterization. To the theory of clusterization, the calculated distance between the two farthest points in the cluster should be less than the distance between the closest two points from different clusters. By this rule, we defined that two parameters of the method should be the maximum distance between points by which these can be connected and assumed to be in one cluster. The second had to be the maximum distance in the cluster, aka the cluster’s diameter. A cluster’s diameter is the farthest distance between two points within a cluster. With these hyperparameters and the defined distance method, we can assume that every point is a vertex of a graph, two points within the threshold of the distance between pairs of ones are neighbours, and count the connection between counts as an edge of a graph. The group of connected vertexes or a particular vertex is a graph. The diameter hyperparameter ought to keep the data homogeneity in a cluster. We can define every graph as a cluster with defined rules based on previous assumptions. Later in this paper will be visualized the clusterization of three-dimensional data points. We took one of the most popular clusterization dataset – the iris dataset for visualizing purposes. The paper contains several examples of clusterization of the dataset with different hyperparameters. We took KMeans [3] as an example of the clusterization method. The method based on BFS is a flexible clusterization method that relies on meta-information about distancing between data points.

3DPCTN: Two 3D Local-Object Point-Cloud-Completion Transformer Networks Based on Self-Attention and Multi-Resolution

Acquiring high-fidelity 3D models from real-world scans is challenging. Existing shape completion-methods are incapable of generating details of objects or learning complex point distributions. To address this problem, we propose two transformer-based point-cloud-completion networks and a coarse-to-fine strategy to extract object shape features by way of self-attention (SA) and multi-resolution (MR), respectively. Specifically, in the first stage, the model extracts incomplete point-cloud features based on self-attention and multi-resolution encoders and predicts the missing partial with a set of parametric surface elements. Then, in the second stage, it merges the coarse-grained prediction with the input point cloud by iterative furthest point sampling (IFPS), to obtain a complete but coarse-grained point cloud. Finally, in the third stage, the complete but coarse point-cloud distribution is improved by a point-refiner network based on a point-cloud transformer (PCT). The results from comparison to state-of-the-art methods and ablation experiments on the ShapeNet-Part dataset both verified the effectiveness of our method.