How to read SemanticKITTI datasets to train my own pointnet2 semseg network

[W090325]

[anassmu]

you can create your own class SemanticKITTIDataset and use it instead of S3DISDataset that provide points and labels, the input of pointnet2 semseg is (batch_size,num_point,9) where default batch_size is 16 and default num_point is 4096 as follow : (x,y,z,r,g,b,normalized_x,normalized_y,normalized_z), you can change normalized coordinates by normals or whatever you see useful for the segmentation.