Closed MeyLavie closed 4 years ago
ground plane height of the nuScenes dataset is quite stable, so gorund plane detection is optional; So I remove it in the latest version of code.
@poodarchu Thank you for the response. So do you say it is hardcoded in the code? I couldn't find where you augmented new "detection" on the ground plane in the code which brought (as you described in your article) the greatest NDS improvement (~8) and that is what I'm looking for.
what changes you made (
git diff
) or what code you wrote uncommentednu_ds.get_sample_ground_plane(root_path, version=version)
under create_data.py in order to create infos_trainval_ground_plane.pklwhat exact command you run: Create data script (generated the above file) and then train for cbgs code (nuscenes dataset).
Expected behavior:
As I read in your paper you use the ground plane in order to augment new bb on it but I couldn't find where you use this information in the code. Found something similar on kitti files but they use some .txt files I couldn't find. In your article, you mentioned that it gave you ~8 nds improvement.
Thank you