2, Download Velodyne point clouds and label data in SemanticKITTI dataset here ... In this case, you might want to train model with smaller quantization grid/ feature map via python ...