Train dataset of semantic segmentation neural network on point clouds

I want to train SqueezeSegV2 deep learning network (or similar). I saw that the train dataset (in https://it.mathworks.com/help/lidar/ug/semantic-segmentation-using-squeezesegv2-network.html) is composed of organized point clouds (M x N x 5) that i able to obtain from my point clouds. The labels are in form of M x N images (.png). How I can get these labels?. I used the Lidar Labeler app and boxLabelDataset with no acceptable results. Thanks.