Lidar or Radar label?

[luluisthebest]

Thank you very much for your work, which has brought us a better platform. I would like to study the 4D radar cubes. Is the label in "rosDS/rslidar_points_clean/" used in your code more accurate or is the autolabelled "radar_gt" more accurate?

[IgnacioRoldan]

There are different things. The rslidar_points_clean are the lidar point clouds, that can be used to train a radar detector. In radar_gt you have the class of each point (like vehicle, pedestrian and so on) that can be used to train a classifier.

[luluisthebest]

There are different things. The rslidar_points_clean are the lidar point clouds, that can be used to train a radar detector. In radar_gt you have the class of each point (like vehicle, pedestrian and so on) that can be used to train a classifier.

Thanks a lot! By the way, I trained the network_noElevation.py and the results indicate overfitting. That's quite strange.

[IgnacioRoldan]

Yes, indeed this is strange. Did you use the default parameters in the repo?