OpenMV H7 RAM usage trained Model

OpenMV Boards
, ,
1

I am using a CNN model created on Edge Impulse on an OpenMV H7. It says that the peak RAM usage of the model is 77.8KB and flash is 619.1KB, but I am using a 32GB microSD card. Even though there is enough RAM to run the model, it keeps giving an error:

image
image782×517 52.7 KB

This is the code i am using :

import ml
import gc 

print("Memória livre:", gc.mem_free())

net = ml.Model('trained.tflite')

image
image1920×1033 94.1 KB

2

When loaded from SD, or any writable filesystem, the whole model, including its weights, is loaded into memory. So you’ll probably need ~700KBs of RAM. Your only option is to embed the model in flash, but currently this requires rebuilding the firmware from source. Later we’ll be adding an option to write models to flash without changing the firmware (it’s called romfs, and it’s a MicroPython feature still in development).

3

Please buy the H7 Plus: OpenMV Cam H7 Plus. It has external SDRAM so it can fit the model.