Improve publish rate for object detection on Jetson TX2, ZED2 Camera, Tegra 3.7.7 SDK with ROS

Hi, I am using ZED ROS (ROS 1) wrapper to conduct object detection (multi-class fast model) on an Nvidia Jetson TX2 with the Tegra 3.7.7 SDK and a ZED2 camera. The issue is that whenever object detection is enabled, the publishing rate for messages is far too low for practical purposes (< 2 Hz as measured using rostopic hz zed2/zed_node/obj_det/objects. Ideally 15 Hz or higher would be acceptable. It seems that the bottleneck might be with Position Tracking, because when tracking is disabled, I can get 30 Hz frequency for topics such as rgb/image_rect_color, but these reduce to < 3 Hz when tracking is enabled by other topics (e.g. by asking for zed2/zed_node/point_cloud/point_cloud_registered), and jetson_clocks seems to not have much of an effect. Object detection seems to always enable some form of positional tracking. Is there any way to increase the rate to get around 15 fps?

I appreciate all the support and am looking forward to your responses.

Update: The issue seems to be fixed when the depth quality setting is changed to 1 (originally was set to 4, NEURAL quality).