How to run FOMO on ESP32-S3 with AI acceleration?

Espressiff claims that using AI acceleration on ESP32-S3 can give 10x better performance in comparison to ordinary ESP32.

EDGE IMPULE claims to have amazing performance with FOMO model.

How can I use both of them together?

I see that there are 3 options to deploy model (Arduino lib, own firmware, and custom c library), which one should I use to utilize full power?

You can use the Arduino library, but It is necessary to replace manually the ESP-NN on the Library the Studio deployed. Look at this tutorial for guidance: TinyML Made Easy: Image Classification -