Hello, I am using a Nova orin dev kit. I wanted to record the image data while doing the mapping with vslam and nvblox so that I could use those images to tune parameters by using the Isaac perceptor later in a PC.
So when launching I am adding the Isaac ros data recorder and Isaac ros h246 encoder along with the Isaac perceptor. Note, I am using the same container component for all these nodes.
Whenever I do this, there is huge loss of frames in the recording and sometimes when I playback the ros bags, I get around 1-2 FPS on the compressed images.
Im not sure which node is causing the frame drops but running just the isaac perceptor, I don’t see any frame drop messages but when I use encoder and recorder, I see the frame drop error. Any help will be really useful. Thanks !!!
Yes, I tried that tutorial and even that has some frame drops but it is negligible.
But what I want is to run isaac perceptor(vslam and nvblox) along with encoder and recorder, such that we do real time tests and also use the rosbag data for later fine tuning some parameters.
I noticed that when the subscription count of the image message increases the frame drops are significantly high and because of this frame drops, the recorded rosbags have very low FPS.
Understood. Recording & encording also impact the performance while running Isaac Perceptor on Nova Orin solely already consumes most compute loading in current release. I forward this request to engineer team.
I did few tests to check the performance and here are the screenshots.
This is when running isaac perceptor along with the encoder and recorder nodes. You can already see the frames drop warning messages. The GPU and CPU usage are high but I still believe it can handle even more nodes.
This is when running only the isaac perceptor( the sensors, nvblox and vslam). I can barely see any frames dropping when running it alone and also less CPU and GPU usage.
We will be closing this topic at this time.
If you have further questions or encounter new issues, please feel free to open a new thread.
Thank you for your understanding.
