Different result between on device and live classification

I’ve trained a model and deployed it on an Arduino Nano ble 33 using the arduino library. I’m getting different classifications when I used the static buffer example with features from the live classification. And run model on browser gets more correct results than continously on-device inferencing.
Example 1: on.249819qh.ingestion-64c87cfdcb-j5c4s

Example 2: on.249819r1.ingestion-64c87cfdcb-7cq9l

Hello @vpeopleonatank,

The live classification uses the float32 version of the model. Did you try with the same version (unoptimized) on your arduino?



1 Like

I 've tried unoptimized version on arduino, the result is consistent now. Thanks for your help.

1 Like