Question/Issue:
Hi, I’m new to the Edge Impulse platform and trying to deploy a simple model for animal activity classification on an Actinius Icarus (nRF9160) board. At the Deployment section of the Edge Impulse interface, I see that the estimation for my quantized model (C++ library) is 15.1 kB of flash memory. I integrated the model on my application (developed using nRF connect SDK and therefore Zephyr), it worked fine but when checking the flash footprint I found out it increased by about 85 kB. I checked the Memory report and found put most of the extra space (around 60 kB) appeared as “No paths”, so I’m not sure where it came from. My build is already configured with the optimize for space option.
So my question is, what does the interface take into account when estimating the flash footprint? Does it leave out, ofr example, feature extraction libraries? Also, is there something I can do to reduce this memory usage (I’ already using EON compiler and quantized model)?
Project ID:
1071877
Context/Use case:
Downloading the C++ library and integrating it on a preexistent nRF connect application.
Steps Taken:
Compile the application including the edge impulse library and run “memory report” from the nRF connect expansion for VSCode
Reproducibility:
I tried with different models and this happens every time.
Environment:
- Platform: Actinius Icarus nRF9160 board
- Build Environment Details: nRF connect SDK