GAPointNet: Graph attention based point neural network for exploiting local feature of point cloud

Abstract Exploiting fine-grained semantic features on point cloud data is still challenging because of its irregular and sparse structure in a non-Euclidean space. In order to represent the local feature for each central that helpful towards better contextual learning, max pooling operation often used highlight most important region. However, all other geometric correlations between corresponding neighbourhood are ignored during operation. To this end, attention mechanism promising capturing node representation graph-based by attending over neighbouring nodes. paper, we propose novel neural network analysis, GAPointNet, which able learn representations embedding graph within stacked Multi-Layer-Perceptron (MLP) layers. Specifically, different weights center efficiently exploit features. We also combine with signature generated our fully extract structures enhance robustness. The proposed GAPointNet architecture tested various benchmark datasets (i.e. ModelNet40, ShapeNet part, S3DIS, KITTI) achieves state-of-the-art performance both shape classification segmentation tasks.