Geometric attentional dynamic graph convolutional neural networks for point cloud analysis

Cui, Yiming; Liu, Xin; Liu, Hong‐Min; Zhang, Jiyong; Zare, Alina; Fan, Bin

doi:10.1016/j.neucom.2020.12.067

Cited by 51 publications

(11 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…These works illustrate that the geometry representation of 3D shapes is important for point cloud understanding. Nevertheless, the authors in [ 28 ] expand the concept of DGCNN [ 8 ] to capture not only the intrinsic features of point cloud but also the extrinsic ones so that the network can learn geometry representations better. Similarly, an EdgeConv based feature fusion method is adopted by [ 29 ], in which an adaptive feature fusion module helps to learn both global and local features.…”

Section: Related Workmentioning

confidence: 99%

ResSANet: Learning Geometric Information for Point Cloud Processing

Zhu

Zhang

Ruan

et al. 2021

Sensors

View full text Add to dashboard Cite

Point clouds with rich local geometric information have potentially huge implications in several applications, especially in areas of robotic manipulation and autonomous driving. However, most point cloud processing methods cannot extract enough geometric features from a raw point cloud, which restricts the performance of their downstream tasks such as point cloud classification, shape retrieval and part segmentation. In this paper, the authors propose a new method where a convolution based on geometric primitives is adopted to accurately represent the elusive shape in the form of a point cloud to fully extract hidden geometric features. The key idea of the proposed approach is building a brand-new convolution net named ResSANet on the basis of geometric primitives to learn hierarchical geometry information. Two different modules are devised in our network, Res-SA and ResSA2, to achieve feature fusion at different levels in ResSANet. This work achieves classification accuracy up to 93.2% on the ModelNet40 dataset and the shape retrieval with an effect of 87.4%. The part segmentation experiment also achieves an accuracy of 83.3% (class mIoU) and 85.3% (instance mIoU) on ShapeNet dataset. It is worth mentioning that the number of parameters in this work is just 1.04M while the network depth is minimal. Experimental results and comparisons with state-of-the-art methods demonstrate that our approach can achieve superior performance.

show abstract

Section: Related Workmentioning

confidence: 99%

ResSANet: Learning Geometric Information for Point Cloud Processing

Zhu

Zhang

Ruan

et al. 2021

Sensors

View full text Add to dashboard Cite

show abstract

“…Además, este tipo de estructuras puede ser fácilmente codificado en cadenas de bits, reduciendo así el tiempo de acceso a las celdas de las mismas. OctNet, por ello, permite trabajar con vóxeles de alta resolución, concretamente, resoluciones superiores a 64 3 . Los resultados obtenidos al trabajar con resoluciones altas son mejores que los obtenidos con resoluciones bajas, siendo los mejores resultados los obtenidos con 64 3 .…”

Section: Octnetunclassified

“…A pesar de ello, el uso de la transformación Χ necesita un conocimiento exhaustivo del operador cuando se pretende aplicar a una red neuronal. Por otro lado, presenta resultados prometedores en segmentación en comparación con métodos como PointNet/PointNet++, SO-Net, DGCNN [3] o SGPN [16]. En clasificación, PointCNN ofrece peores resultados con respecto a otros métodos.…”

Section: Pointcnnunclassified

“…Yue Wang et al [3] introducen dos conceptos interesantes para las redes neuronales de grafos (Graph Neural Networks, GNN). Por un lado, aplican una novedosa operación denominada EdgeConv, con el que se consigue obtener las características locales de los puntos, y la relación de estos con sus puntos más cercanos, mientras se respeta la invariancia a permutaciones.…”

Section: Dgcnnunclassified

See 1 more Smart Citation

Análisis comparativo de segmentación semántica de nubes de puntos con redes neuronales.

Espacio¹,

Salamanca²,

Merchán³

et al. 2021

Xlii Jornadas De Automática : Libro De Actas

View full text Add to dashboard Cite

Desde sus inicios, el ámbito de la visión por computador se ha esforzado por recabar de imágenes 2D y espacios 3D la información y el conocimiento que el ser humano puede extraer fácilmente con un solo vistazo. Los métodos más efectivos han apostado por llevar el concepto de inteligencia artificial un paso más allá, optando por unir la extracción de información con el uso de redes neuronales, dando como resultado lo que hoy llamamos aprendizaje profundo, una de las ramas de la computación que más desafíos ofrece a día de hoy. Este artículo no busca sino realizar una recopilación de aquellos métodos que siguen siendo la base, como PointNet, de muchos estudios recientes, así como aquellos que bien podrían suponer nuevas líneas de investigación en el campo de la segmentación semántica de escenas y modelos tridimensionales, dada la importancia y el reto derivado de trabajar con nubes de puntos. Con este objetivo, se realiza una comparativa entre los métodos para quienes deseen introducirse en este campo, considerando aspectos que se creen convenientes para trabajar con nubes de puntos reales.

show abstract

“…However, it lacked the ability to extract local information, and it was inappropriate to extract the nearest neighbors under the uneven density of point cloud. The point cloud learning networks [32][33][34] under normalized input strongly depend on data sources. The point cloud is affected by the acquisition equipment and the coordinate system, and its arrangement is changeable.…”

mentioning

confidence: 99%

3D Instance Segmentation and Object Detection Framework Based on the Fusion of Lidar Remote Sensing and Optical Image Sensing

Bai

Cen

et al. 2021

Remote Sensing

View full text Add to dashboard Cite

Since single sensor and high-density point cloud data processing have certain direct processing limitations in urban traffic scenarios, this paper proposes a 3D instance segmentation and object detection framework for urban transportation scenes based on the fusion of Lidar remote sensing technology and optical image sensing technology. Firstly, multi-source and multi-mode data pre-fusion and alignment of Lidar and camera sensor data are effectively carried out, and then a unique and innovative network of stereo regional proposal selective search-driven DAGNN is constructed. Finally, using the multi-dimensional information interaction, three-dimensional point clouds with multi-features and unique concave-convex geometric characteristics are instance over-segmented and clustered by the hypervoxel storage in the remarkable octree and growing voxels. Finally, the positioning and semantic information of significant 3D object detection in this paper are visualized by multi-dimensional mapping of the boundary box. The experimental results validate the effectiveness of the proposed framework with excellent feedback for small objects, object stacking, and object occlusion. It can be a remediable or alternative plan to a single sensor and provide an essential theoretical and application basis for remote sensing, autonomous driving, environment modeling, autonomous navigation, and path planning under the V2X intelligent network space– ground integration in the future.

show abstract

Geometric attentional dynamic graph convolutional neural networks for point cloud analysis

Cited by 51 publications

References 6 publications

ResSANet: Learning Geometric Information for Point Cloud Processing

ResSANet: Learning Geometric Information for Point Cloud Processing

Análisis comparativo de segmentación semántica de nubes de puntos con redes neuronales.

3D Instance Segmentation and Object Detection Framework Based on the Fusion of Lidar Remote Sensing and Optical Image Sensing

Contact Info

Product

Resources

About