deepstream-app with tracker or not have large difference in Frame throughput？

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

[b]i test deepstream-app with tracker or not but have large difference in Frame throughput
the result as follows:

stream number	witch tracker	witchout tracker	difference
1        	112.46	         132.4	                 19.94
2	        100.53	         125.89	                 25.36
4	        81.6	         116.49	                 34.89
8	        56.64	         98.62	                 41.98
16	        31.63	         63.61	                 31.98
20	        25.38	         51.75	                 26.37
32	        15.27	         34.17	                 18.9

1.why with tracker or not has a great influence，my config as fllows，is that right？

2.frame number double but Frame throughput not in half，Is the reason of streammux plugin?

streammux send batched GstBuffers and nvinfer handles them in batched[/b]

2-witch tracker.txt (3.94 KB)

2-witchout tracker.txt (3.94 KB)

4-witch tracker.txt (3.94 KB)

4-witchout tracker.txt (3.94 KB)

witch tracker or not.html (6 KB)

double input stream.html (13.1 KB)

2

Can you change the config Primary-gie → interval to be 1, and test again.

If interval = 0, primary-gie will do inference for every frame.
If interval = 1, primary-gie will do inference in every 2 frames, tracker will get bbox for the other frame
If interval = 2, primary-gie will do inference in every 3 frames, tracker will get bbox for the other 2 frames.

frame number double but Frame throughput not in half

  1. Decoder is no the bottle neck
  2. The performance of inference for batchSize=4 (4 frames) is NOT much worse then the performance of inference for batchSize=2 (2 frames)