How to reach 455[FPS] with peoplenet sample on Orin DevKit


I can succeed to run peoplenet sample with JetPack5.0.2 and Deepstream6.1.1 on Jetson OrinDevkit like below command.

/opt/nvidia/deepstream/deepstream-6.1/samples/configs/tao_pretrained_models$ deepstream-app -c deepstream_app_source1_peoplenet.txt

And we also found below performance table.

Our result is like below;

  • We modified config file like below.

  • We found 271[FPS] on OrinDevKit(MAXN setting) like below;

Are there anyone success to run the peoplenet sample near 455[FPS] ?

We would appreciate if you tell us the proper way to reach 455[FPS] with OrinDevKit.

Best regards,

Did you boost CPU/GPU/EMC to max clock?
sudo nvpmodel -m 0
sudo jetson_clocks
Did you turn off osd and tiler?
about tracker setting, did you follow part: Use the max_perf setting for tracker
The numbers are obtained with sample_1080p_h265.mp4.

Thank you very much for your kindness.
I really appreciate your help.

I tried to modify the configuration based on your guide.
Then, We found about 360[FPS] .
Thank you!
But We cannot find near 455[FPS].

Is there any other way to increase FPS?

I’d like to share our config files.
So, I would appreciate if you check the behavior with your OrinDevKit.

Best regards,

deepstream_app_source1_peoplenet.txt (3.5 KB)
config_infer_primary_peoplenet.txt (2.4 KB)

There is no update from you for a period, assuming this is not an issue anymore.
Hence we are closing this topic. If need further support, please open a new one.

This performance is expected due to a heavier model. We will keep updating to newer models from time to time and adding performance for older ones may be dropped.

Hi @mat-h ,
Form your configure file, you are using batch size = 1, you need use larger batch, e.g. 16 or 32, for higher FPS.

Thank you very much for your kindness.
I really appreciate your help.

I’d like to check the behavior from your guide.

Best regards,

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.