Text this: Finding the Optimal Convolutional Kernel Size for Semantic Segmentation of Pole-like Objects in Lidar Point Clouds