Multiscale Receptive Fields Graph Attention Network for Point Cloud Classification

Figure 5

The architecture of classification. In this framework, it takes points as input and applies individual SRFGAT modules to obtain multiattention features on multilocal graphs; then, the output features are recast by means of five shared MLP layers and attention pooling layer, respectively. Finally, a shared full-connected layer is employed to form a global feature and then classification scores for c categories are obtained.