Fig. 12
From: Frame points attention convolution for deep learning on point cloud
FPA can also generate continuous and smooth filter weights. The general surface features of the lampshade or bed have small differences, although quite different frame weights, FPA can still generate a spatially smooth convolution filter.
