Fig. 4
From: Frame points attention convolution for deep learning on point cloud
The convolution of the local area of the point cloud is more slimier to the equivalent form of 2D regular data convolution. The key issue is how to define \(\widetilde{g}(x_n)\).
