How to allocate more memory for IA models

[sisimonis 5 1 1]

Hello, I'm currently working on image classification model on spresense.

I managed to put it on the card but on the execution i face the classical :
Arena size is too small for all buffers. Needed 1215024 but only 270336 was available.

I know how to deal with it, by increasing the allocation here :
constexpr int kTensorArenaSize = 300 * 1024;

But my problem is that i can't increase it higher than approximately 650*1024. I'm aware the default value of ram size allocated for memory is 760 KB (as in Arduino). And in Arduino, i managed to increase it and so my model can fit in the card. But with the sdk i'm stucked.

I modified person_detection example from tflite to put mine.

Any ideas ?

Thanks
Simon

[CamilaSouza]

This post is deleted!

[CamilaSouza]

This post is deleted!

[sisimonis 5 1 1]

Something wrong in my question ?

[CamilaSouza]

@sisimonis-5-1-1

No, my bad. Don't worry.
I'll get back to you as soon as possible.

[CamilaSouza]

Hey, @sisimonis-5-1-1

The thing is.. In the SDK, unless you have ASMP enabled, the default value of RAM size is 1536KB.
You can check if you have the ASMP config enabled and disable that. Or if you need it, then you can control the ASMP shared memory size with the ASMP_MEMSIZE config.