Open nikolauswagner opened 4 years ago
This issue could likely be addressed by using gazebo services. Since we are using mostly rosbags at the moment (which obviously do not provide those), this has to be addressed in such a way that it can use these services to refine the bboxes when the services are available, but uses a coarse-grained approach when run on a rosbag.
Bounding boxes are currently rough estimates based on approximate object sizes and known distance to the camera. Make them more accurate by e.g. making use of the orientation / collision model / ...