What causes large latency jitters when SGIEs are added?

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU): Jetson AGX Xavier
• DeepStream Version: 5.0
• JetPack Version (valid for Jetson only): 4.4
• TensorRT Version: 7.1
• Issue Type( questions, new requirements, bugs): questions

I have the following pipeline:

         ---> muxer -> pgie -> tracker -> sgie1 -> sgie2 -> app_sink 

The app_sink call a function that trigger external hardware based on the result of of sgie1 & sgie2 on detected object. We measure the latency from the time the images were captured at the camera to the time the external hardware receive the instruction.

When sgie1 & sgie2 are included, the latency range from 68ms to 80ms, however, when we removed sgie1 & sgie2 the latency does not change as much and range from 50-53ms. What could be the cause of large latency jitters when SGIEs are added?

Can you set CPU/GPU/EMC clock to constant, like to max,
sudo nvpmodel -m 0
sudo jetson_clocks
and try again?

It’s expected since the sgie perform inference on the result from pgie, that means if the pgie detect more objects, then sgie will consume more time to do inference.

1 Like