Abstract: Kernel point convolution (KPConv) defines convolutional weights based on Euclidean distances between kernel points and input points and has shown good segmentation results on several ...