When saving point clouds for object and then using those to find objects again we use the same resolution (downsampling size) as the tracker. However this doesn't work that well for smaller objects. So we need to use higher resolution for saving and finding objects (for accuracy) and then switch back to the lower resolution for tracking (for speed).
May be an idea to split the node into two, one for tracking and one for finding objects. Will need an interface on the tracker that we can pump in clouds to track with.
When saving point clouds for object and then using those to find objects again we use the same resolution (downsampling size) as the tracker. However this doesn't work that well for smaller objects. So we need to use higher resolution for saving and finding objects (for accuracy) and then switch back to the lower resolution for tracking (for speed).