Amount of Memory needed for various Networks?

I'm looking to deploy a trained network onto an ARM chip but limited by memory. Is there a way to figure out how much memory is needed or compute availablility is needed for a specific network?

Answer 1

[Ranjeet]

Hi Wilson,

The amount of memory needed by a network depends on number of learnable parameters and activations. You may analyse a network using the following network analyser functionality –

Analyze a network

Network analyser provides data for a network like the following image –

Total learnable and activations can be summed up and multiplies by the data type that is used to store to the network size. Data type can be ‘float32’, ‘int16’, ‘int8’, etc.

Also, you may find the following resource useful w.r.t deploying a network on ARM microcontroller –

How to Deploy Machine Learning Model to ARM Cortex -32 bit Microcontroller?