MemorryError -H7-48x48 impuls

I’m doing my first steps with Image recognation with OpenMV. And i faced with a problem of Memory error: Out of Fast Frame Buffer Stack Memory.
I have Cam H7, and i’m using 48x48 pixel input network (even tryed 36x36), but still i got this error. Execuse me if anyone on this forum already had the same issue and you already replyed.
Can this problem be connacted with a fact, that i’m using 128gb sd card?

Thank You!

Can you reduce the image your are capturing? Like, set your res to QQVGA?

You are meaning the resolution of those images on which i trained my network?

Still the same error:( even if i’m using QQVGA images for training.

Mmm, well, the H7 really doesn’t have a lot of RAM. How big is your network? And what’s it doing?

A lot of folks don’t realize that you get a copy of the previous layer in the network per convolution. So, if you have a x64 convolution layer with a 48x48 image this requires 64 * 48x48 pixels. Then if you convolve that again without a pool it grows even bigger.

So, it’s possible to have a network with only a few weights needs a very large amount of buffer space.

Hmmm, okey
So what settings in the Edge Impulse for network training for H7 you recommend?

Thank You!

Okey, i just reduced the neurons number and now it works. But network not so accuarate, maybe i will think about H7 Plus camera for better perfomance.

Thank You!