[CVPR 2023] TPVFormer: Tri-Perspective View for Vision-Based 3D Semantic Occupancy Prediction

An academic alternative to Tesla's occupancy network for autonomous driving. #CVPR #CVPR23 #autonomousdriving

[@chriscohoat]

Great work!

[@XYY-hi9uy]

Actually, I have a question regarding this 3D occupancy prediction task: how is it related to the semantic segmentation task? For instance, if I use a semantic segmentation network to assign labels to points in a point cloud in real-time, can I then simply proceed to voxelization? Could I determine the label for each voxel based on the labels of the points within that voxel, and then output occupancy information?