Isaac ROS integration with existing DeepStream pipeline for DNN video inference


We currently have a DeepStream pipeline that we use to ingest video data and perform inference using a Yolov5 model converted to TensorRT. Isaac ROS and NITROS have parallel capability, but what about the case where you want to leverage both Nvidia solutions?

My understanding is the idea behind DeepStream is to also avoid unnecessary GPU/CPU memory copies and leverage GPU/hardware acceleration as much as possible. Seems like a very similar concept as NITROS, no? If we already have an accelerated pipeline that leverages DeepStream, what do you advise is the best path to add NITROS support?

Hi @percolate ,

At this time, we are considering adding DeepStream to our roadmap.


Good to know Raffaello. From a user endpoint, it would be great to be able to easily leverage NVIDIA’s multiple frameworks, so this is good news :)

So what is currently the canonical way to get an image into nitros? for cameras I see there is an Isaac ROS argus camera node, but for a video, I only see the ros isaac compression node (h264 encoding/decoding), not anything that reads a video file or a live video stream…how do you do that?