Failed to allocate TFLite arena (error code 1) in STM32CubeIDE

[ang_anz]

Hello, I’m trying to deploy my image classification EI project on a STM32L475VG-IOT01A board. I downloaded the Cube.MX CMSIS-PACK from EI and followed the tutorial provided by Edge Impulse (https://docs.edgeimpulse.com/docs/using-cubeai). The only thing I changed from the provided code is the following:

---

static const long features[] = {0xffffff, 0xffffff, …, 0xffffff, 0xffffff, 0xffffff};

int get_feature_data(size_t offset, size_t length, float *out_ptr) {
memcpy(out_ptr, features + offset, length * sizeof(long));
return 0;
}

---

Actually I only changed the features type from float to long(but the problem is present also with the float type). When I build the project I have 0 errors and 0 warnings, but when I try to run I got this on Putty:

Anyone could help me with this problem?

Thanks,

Angelo

[aurel]

Hi Angelo,

The error means we can’t allocate enough memory to load the NN, the ST board is limited in RAM. Which MobileNet model are you using?
I would suggest trying with one of those 2 MobileNet v1 models as they require less RAM:

Best,
Aurelien

[ang_anz]

Thanks Aurelien, the peak RAM usage of the NN (115KB) was slightly less than the RAM on the board (128KB); I tried with MobileNetV1 0.2 with a smaller NN and it’s working well!

[janjongboom]

@ang_anz Yeah you’ll lose some static RAM here and there, f.e. stack so you’ll have a bit less RAM available than the full 128K.