Hi everyone!
I’ve been diving deep into Edge AI since some months, and honestly, the progress we’re seeing on ARM-based IoT is mind-blowing. moving a model from a powerful PC to a tiny Cortex-M controller is where the real 'fun' (and headaches!) begins.
I’m curious to know what everyone is using these days to keep their models lean and mean. I've been playing around with TensorFlow Lite Micro, but I keep hitting walls with memory overhead.
A few things I’d love to chat about some points. Is anyone else finding that ARM’s CMSIS-NN gives that extra 'oomph' in speed over standard kernels? Do you guys find the jump from 16-bit to 8-bit quantization worth the slight drop in accuracy for real-world sensor data? Is Edge Impulse still the king for quick prototyping, or has a new 'hidden gem' tool popped up on your radar?
I’d love to hear about your latest wins (or even the projects that crashed and burned!). Let’s share some insights and maybe save each other some debugging time.