Thanks for your interesting work!
I have a question about the generation of gt labels.
The Kitti dataset has already labeled the groundtruth on BEV and provided the position for objects. Why do you regenerate the labels by point cloud completion and hight compression? Why not just use the bev labels provided by the dataset?
Thanks for your interesting work! I have a question about the generation of gt labels. The Kitti dataset has already labeled the groundtruth on BEV and provided the position for objects. Why do you regenerate the labels by point cloud completion and hight compression? Why not just use the bev labels provided by the dataset?