Hi everyone,
I am designing an object detection and tracking pipeline. Is it possible to use/integrate/steal an object tracker (like NvDCF) from deepstream within isaac framework?
OR is there already a Isaac-friendly alternative? I see plenty of examples for the upper part of pipeline (image pipeline + object detection, e.g. Isaac ROS Object Detection — isaac_ros_docs documentation ) but nothing for the tracking that typically come after.
Thanks in advance!
Welcome to the Isaac ROS forum!
Have you check also Isaac ROS Pose Estimation — isaac_ros_docs documentation ?
Best,
Raffaello
Hi!
Although what is in the isaac ros pose estimation looks very nice, the three models implemented there really focus on pose estimation while tracking seems more like a byproduct… As a matter of fact all three options requires some level of training information of the object to be estimated/tracked while NvDCF and similar trackers (e.g. SORT based multiple object tracker) do not need that, which simplify a lot its usage!
I am actually interested in the more basic functionality of frame-to-frame tracking of bounding boxes with re-identification, and I saw few options in other nvidia frameworks (e.g., Deepstream and NvDCF) but they do not seem to be usable with (isaac) ros eco-system. Are there any way/plan to unify the different frameworks? Or are there already options out there for object tracking?