Lecture Description:

In this talk, you will learn how we implemented and optimize a 3D point cloud based neural network to detect and segment an input scene. Directly taking LIDAR data at 3D without converting to 2D images has significant saving in compute but also preserves valuable accuracy of the model.