cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
deepg799
Explorer
Explorer
374 Views
Registered: ‎01-20-2019

Vitis AI samples - performance miss matching

I have measured the performance number of the individual Vitis-AI samples application such as facedetect, classification, and so on by using the Vitis_AI samples respective performance benchmark application.

example - facedetect

- test_performance_facedetect densebox_640x360

Below are my observation and question with respect to the Vitis-AI sample's performance

1 - How exactly test_performance_facedetect densebox_640x360 is measuring the performance number because of the inference performance number of the sample application  "test_video_facedetect densebox_640x360" (running on video file 640x360) almost 2x time low as compare to the number which I am getting from the test_performance_facedetect.

2Why the Vitis-AI samples application giving me the same performance number even if I am increasing the DPU core number to two (4096) from one (4096)? It seems the sample application is not utilizing the second core of DPU

3 - Any suggestion to optimize and utilize the DPU cores efficiently?

Thank you in advance.

Tags (4)
0 Kudos
Reply
1 Reply
jasonwu
Moderator
Moderator
260 Views
Registered: ‎03-27-2013

Hi @deepg799 ,

 

If you are using video as input resource the input throughput is limited by video stream bandwidth so that you may not have DPU run the a best performance.

Not sure if it is the issue you met?

 

Best Regards,
Jason
-----------------------------------------------------------------------------------------------
Please mark the Answer as "Accept as solution" if the information provided is helpful.

Give Kudos to a post which you think is helpful and reply oriented.
-----------------------------------------------------------------------------------------------
0 Kudos
Reply