No acceleration than Nano

Hi

We found NX performance was not better than Nano as running FastMOT model.
Is there any way to improve the performance with FastMOT on NX?

GitHub - GeekAlexis/FastMOT: High-performance multiple object tracking based on YOLO, Deep SORT, and optical flow

Thank you,

Thanks for reporting this issue.
Let us reproduce it in our environment first. Will get back to you later.

Thanks.

Hi,

Sorry that we cannot reproduce this issue on JetPack4.5.
Could you also give it a check?

We test FastMOT with video mode and XavierNX achieves more than twice average FPS as below:

  • Nano: 11 (fps)
  • Xavier-NX: 25 (fps)
$ sudo nvpmodel -m 0
$ sudo jetson_clocks
$ python3 app.py --input_uri /opt/nvidia/deepstream/deepstream-5.0/samples/streams/sample_720p.mp4 --mot --gu

Thanks.

Hi AastaLLL,

Thank you for your prompt support.

Yes. As nvpmodel is set to 0, NX performance achieves more than twice average FPS.
Two more questions before we do fully test:

  1. Is there any special reason that FastMOT got better performance as running with two cpu than six cpu? Doesn’t FastMOT use cpu more?

  2. On NX system we also use some other applications which use CPU, could it cause other performance issue as only use two cpu of NX?

Thank you,

Hi,

1.
You can find the power mode details below:

https://docs.nvidia.com/jetson/l4t/index.html#page/Tegra%20Linux%20Driver%20Package%20Development%20Guide/power_management_jetson_xavier.html#wwpID0E0VO0HA

The sample runs FastMot with TensorRT, which is deployed on the GPU.
So the number of CPUs doesn’t have too much impact on the performance.
But the higher frequency in mode 0 can help to run the sample faster.

2
Actually, if power budget is not a concern, you can enable them manually.

Thanks.

Hi AastaLLL,

Thank you for the information.
We will test it.

Hi AastaLLL,
When i use the same FastMot, i get the 10 FPS in Jetson Xavier AGX with max clock and power, i do not know the problem !!. can you tell me how i increase FastMot. please

Hi ouziane.mhm,

Please help to open a new topic. Thanks