Different inference time

Help
#1

Hi everyone,
I’m new to Edge Impulse. I have created my simple project and faced anomaly scenario that the inference time in the studio is different from when deploying in the hardware. In the studio, it showed that about 1120ms while in my case, when running on the board, the inference time approximately about 3065ms.

image
image755×499 106 KB

Can anyone help me explain this issue clearly and let me know how to decrease the inference time when deploying in the board?

Model: Image classification using transfer learning against MobilenetV1
Target: STM32F407 DIscovery running at 168 MHz, 192Kb RAM and 1Mb Flash
Software: STM32CubeIDE

#2

Hi @trieuhuynh110600,

There could be many things going on that affect inference time on the board vs. simulation. What target board do you have selected in your project? Alternatively, if you post your project ID, I can also check.

#3

Hi @shawn_edgeimpulse,

My project ID is 383255. My target board is STM32F407 Discovery which running against Cortex M4. In addition, I selected the Cortex M4F option since I did not see any option for my board. It would be highly appreciated if you could take a look and give your comments. Many thanks in advance.