November 24, 2020, 7:38am
I tried to deploy different models " mainly image classification " as STM32 CMSIS pack , but It’s not working.
I saw in another topic that this issue is related to the version of Cube.Ai in use ( 5.0.0).
Since Cube.Ai v 5.2.0 was released … Is this version will solve the problem?
If yes , when can we expect the support of the new cube.ai version and the fix of the cmsis pack build?
@mc1994, apparently 5.2.0 would indeed solve it, but the structure of Cube.AI has changed and we’d need to revalidate it.
In the meantime you can use
Build C++ Library with STM32CubeIDE to use the C++ Library export from your CubeIDE project.
November 24, 2020, 11:37am
Thanks for the details .
The deployment using C++ Library with STM32 is working just fine for me, but I’m waiting for a better performance using the cmsis pack + cube.ai 5.2.0 option.
Waiting for your updtaes , keep rocking guys
@mc1994 great. Note that:
but I’m waiting for a better performance using the cmsis pack + cube.ai 5.2.0 option.
I don’t think you’ll get better performance as long as you deploy using EON. EON uses similar kernels underneath so latency will be similar, and EON does some other tricks to lower memory usage compared to CubeAI.
CubeAI is useful if you need a C environment though.
November 24, 2020, 12:41pm
Ah , so The cmsis pack / cube.ai deployment option will not use cube.ai run time library ?
It will, it’s just probably similar in speed than the C++ Library export as they both use CMSIS-NN kernels underneath.