I’m working on (animal) sound detection & considering training the (CNN) model using Keras on a desktop & then importing into EI.
I’m wondering if anyone has guidelines on the max complexity (layers etc) for the model? It’ll be run on an ESP32.
Complexity is highly subjective and max layers can mean many different things, especially if you change the number of nodes per layer. I highly recommend setting your target device to the ESP32 in your Edge Impulse project and checking the estimated RAM/ROM usage and inference time on your Deployment page after you train. That will give you an idea if your model will fit on the ESP32.
Thanks for the response. My question was quite vague so apologies!
If anyone else is viewing this this is a link to some example projects which may be useful in getting an idea of possible complexity & size.