Spatial Aggregation Net: Point Cloud Semantic Segmentation Based on Multi-Directional Convolution

Cai, Guorong; Jiang, Zuning; Wang, Zongyue; Huang, Shangfeng; Chen, Kai; Ge, Xiaojia; Wu, Yundong

doi:10.3390/s19194329

Cited by 11 publications

(16 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…An accuracy of 0.8153 was achieved by the proposed method on S3DIS. This was three percentage points higher than that of the latest spatial aggregation net (SAN [5]). Our method achieved better results than those of PointNet [2], PointSIFT [11], KCNet [12], and SO-Net [13].…”

Section: Introductionmentioning

confidence: 61%

“…However, the extraction and use of local features still face great challenges. Excellent achievements have been obtained by the latest multidirectional convolution network (SAN [5]) combined with spatial structure information in semantic segmentation of point clouds. However, SAN cannot avoid loss of information during subsampling, the regional network is not normalized, samples are not assigned to a certain number of points, there is edge effect, the segmentation on large-scale data sets takes a long time.…”

Section: B the Methods Of Deep Learningmentioning

confidence: 99%

“…1, the nth sample point is represented by N. Let the numbers of input and output feature maps be L and H respectively. The local domain k of a point is searched by a KK octant search, the octant search [5] is first done by a KK octant search, and the result is further selected by K-NN by setting a threshold (K). The mth neighbor is represented by M. The local features obtained by searching are represented by F, F ∈ R N×M×l .…”

Section: A Multi-scale Feature Framework(mff)mentioning

confidence: 99%

“…Spatial and geometric features are ignored.In order to solve the irregularity and disorder of the point cloud, coordinate convolution [3] can effectively extract local information. The point cloud surface structure is damaged by the downsampling of the voxel grid.Since the advanced geometric correlation between the input and its neighboring coordinates and features is not fully utilized,suboptimal segmentation performance is achieved [4].There are errors in segmentation of multi-connected regions.The input of the network should be normalized to the specified number of points,information is lost by the sampling process [5]. The latest research shows that segmentation performance can be improved by using a multiscale feature fusion method [5], there are still some problems in this scheme: the regional network is not normalized, samples are not assigned to a certain number of points, there is edge effect, the segmentation on large-scale data sets takes a long time.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

MLFNet- Point Cloud Semantic Segmentation Convolution Network Based on Multi-Scale Feature Fusion

et al. 2021

View full text Add to dashboard Cite

The project was funded by jinan science and technology bureau and undertaken by QiluUniversity of Technology(Shandong Academy of Sciences),Machine vision-based online intelligentsegmentation of foreign objects in liquids is implemented.2019GXRC067. ABSTRACTIn the semantic segmentation of a point cloud, if the spatial structure correlation between the input features and coordinates are not fully considered, a semantic segmentation error can occur. We propose a method of spatial convolution that makes full use of the characteristics of a multiscale spatial structure by combining local and global features. We call this method MLFNet. We also propose a multiscale feature framework. First, the point cloud is simplified by obtaining the weighted farthest point (by down-sampling combined with farthest-point sampling and the weighted average). The near-near domain of each sampling point is then obtained by a KK octant search (an octant search optimized by the knearest neighbor and a custom threshold), and feature information is obtained. The feature information is added to the subsequent multilayer perceptron, and fusion of local context information is achieved. Finally, the fusion features in multiple directions are maximally pooled. Our method was tested on self-made datasets and other standard basic datasets (ModelNet40, ShapeNet, and Stanford large-scale 3D indoor spaces (S3DIS) data). The accuracy of segmentation was 0.937 in our dataset; two percentage points higher than the latest deep learning method. Also, our method obtained a mean intersection over a union of 0.867 in ShapeNet, which was 0.3 percentage points higher than the latest PointGrid. The accuracy on S3DIS was 0.8153, which was three percentage points higher than the latest spatial aggregation net. The results of semantic segmentation verified the superiority of the proposed method.

show abstract

Section: Introductionmentioning

confidence: 61%

Section: B the Methods Of Deep Learningmentioning

confidence: 99%

Section: A Multi-scale Feature Framework(mff)mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

MLFNet- Point Cloud Semantic Segmentation Convolution Network Based on Multi-Scale Feature Fusion

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Big data from different sensors are the basis for the Internet of Things (IoT) to play its own advantages. Point cloud is an important 3D data format that is widely used in 3D semantic segmentation [ 1 , 2 ] and 3D object detection [ 3 , 4 ]. Moreover, point cloud is a raw output format of many 3D capturing devices such as depth cameras and LiDAR sensors.…”

Section: Introductionmentioning

confidence: 99%

A Novel Point Cloud Encoding Method Based on Local Information for 3D Classification and Segmentation

Song¹,

Gao²,

Li³

et al. 2020

Sensors

View full text Add to dashboard Cite

Deep learning is robust to the perturbation of a point cloud, which is an important data form in the Internet of Things. However, it cannot effectively capture the local information of the point cloud and recognize the fine-grained features of an object. Different levels of features in the deep learning network are integrated to obtain local information, but this strategy increases network complexity. This paper proposes an effective point cloud encoding method that facilitates the deep learning network to utilize the local information. An axis-aligned cube is used to search for a local region that represents the local information. All of the points in the local region are available to construct the feature representation of each point. These feature representations are then input to a deep learning network. Two well-known datasets, ModelNet40 shape classification benchmark and Stanford 3D Indoor Semantics Dataset, are used to test the performance of the proposed method. Compared with other methods with complicated structures, the proposed method with only a simple deep learning network, can achieve a higher accuracy in 3D object classification and semantic segmentation.

show abstract

Bioinspired Multifunctional E‐skin for Robot Dynamic Tactile Real‐Time Feedback Systems Using Triboelectric Sensors and Electrochromic Devices

Wang

Fang

Wang

et al. 2022

Advanced Sensor Research

View full text Add to dashboard Cite

Designing efficient robot dynamic tactile feedback systems remains a great challenge, especially in disaster relief and human-computer interaction, because they need to provide timely tactile and visual data. Herein, a novel bioinspired electronic skin (e-skin) is proposed based on self-powered triboelectric sensor (TES) arrays and electrochromic device (ECD) arrays. Single-electrode TES arrays (TESA) are used to detect pressure distribution at the different positions of the robot. The ECD arrays (ECDs) are used as the real-time visualization window to provide feedback on the pressure distribution by the change in the color of the e-skin.The bioinspired e-skin arrayed system possesses a similar capability to Chameleon, displaying the pressure information on the contact surface between the robot and object in real-time via interactive color-changing capabilities, such as the location, degree, and distribution. Moreover, the proposed e-skin can be used as a feedback system for large-scale curved robot surfaces due to its flexibility, large-scale process, and excellent compatibility. Thus, the novel design, large-scale process, and self-powered advantages make it a good candidate to pave the way for developing human-robot interactions as robot dynamic tactile real-time feedback systems.

show abstract

Spatial Aggregation Net: Point Cloud Semantic Segmentation Based on Multi-Directional Convolution

Cited by 11 publications

References 25 publications

MLFNet- Point Cloud Semantic Segmentation Convolution Network Based on Multi-Scale Feature Fusion

MLFNet- Point Cloud Semantic Segmentation Convolution Network Based on Multi-Scale Feature Fusion

A Novel Point Cloud Encoding Method Based on Local Information for 3D Classification and Segmentation

Bioinspired Multifunctional E‐skin for Robot Dynamic Tactile Real‐Time Feedback Systems Using Triboelectric Sensors and Electrochromic Devices

Contact Info

Product

Resources

About