• Hardware Platform (Jetson / GPU) V100
• DeepStream Version 5.0 Developer Preview
• TensorRT Version 188.8.131.52
I’ve been building off of the sample
deepstream-app (YOLOv3 object detector) to familiarize myself with the SDK.
Ultimately, I’m trying to do the following:
- Grab a system timestamp when a frame from a live video stream (maybe a RTSP source or V4L2 for a USB camera) arrives at the machine. This timestamp value is not used in the inference process nor given to the object detector, but I want to associate it with the output from the inference process for that specific frame, so I need to save it until inference finishes.
- Create a single network socket when the app starts up, and pass that socket in so that I can use it to send messages containing both inference output for a frame (bounding boxes, tracker IDs, etc.) and the aforementioned timestamp of that specific frame.
Basically I’m trying to figure out where in the code I should be grabbing this timestamp, where I should be creating my socket (in
deepstream_app_main.c?), and how I should be passing the socket and timestamps through with the app so I can use them in other methods once the object detector and tracker have finished doing their thing.
deepstream_app.c I see that
write_kitti_track_output(), which I could modify to send socket messages instead of printing stuff to a file, but I can’t create the socket in there as it would end up getting created and destroyed for every frame of video.