How to run 4 python object detections from 4 sources at the same time and display in one window?

[AK51]

Hi,

I have a AGX and want to process 4 video sources in parallel and display them in one windows (2x2)
I can run dusty object detection perfectly. But in python, I need to process each source in sequence which slow down the loop a lot I guess. Threading? I tried but it is slow (may be I did something wrong, it there any example?)
I saw there is some cuda synchronization I can do in python, is it true?
Plan B, I run 4 process and output the rtsp, and use another program to accept all 4 rtsp and display them in a single window. But I am worried about memory usage and speed also.

Thx

[kayccc]

Is issue on AGX Xavier or AGX Orin?
Which JetPack version?

Thanks

[AK51]

AGX Xavier.
Thx

[DaneLLL]

Hi,
For this use-case you can try DeepStream SDK and use nvmultistreamtiler. Please download and try this sample:
deepstream_python_apps/apps/deepstream-test3 at master · NVIDIA-AI-IOT/deepstream_python_apps · GitHub

[AK51]

Hi,

I have terrible experience of installing deep-stream…
Is there a proper instruction of installing 6.1? And run your example?
I could only found 5.0.1 installation guide…
https://docs.nvidia.com/metropolis/deepstream/5.0.1/dev-guide/text/DS_Quickstart.html
Oh, I was looking for posenet instead of object detection I mean.
And it loop back what I did :<, Any complete installation guide for “deepstream_pose_estimation”? - Intelligent Video Analytics / DeepStream SDK - NVIDIA Developer Forums

Thanks.