Conv2D x Conv1D

Hi!

I am currently working with point cloud pose regression and classification and I implemented PointNet as the basis for my architecture.

My question is: why use 2D convolutions instead of 1D convolutions? What are the reasons and advantages for that approach?

Thanks in advance!