In this demonstration, we show how two CNNs trained on different tasks can be sent via BLE to a Nicla Vision MCU (Cortex-M7) and then executed on the device. This can be done without flashing or resetting the MCU at runtime by dynamically reconfiguring the CNN running on the MCU based on the information received via BLE. Both the topography of the CNN and its weights can be adjusted this way within in the resource constraints of the MCU.
This demonstration shows a deep neural network running on an Arduino Nano 33 BLE Sense microcontroller performing keyword spotting. The DNN can distinguish 36 different spoken keywords (Pete Warden’s Speech Commands dataset), infers at about 800ms per 1 second audio sample, and requires 190KB of ROM and 111KB of RAM.
Demo of on-device DNN training of an animal classification dataset with 10 classes on a STM32H743VI ARM Cortex-M7 microcontroller (OpenMV H7 R2).
Demo of MicroYOLO person detection on a STM32H743VI ARM Cortex-M7 based microcontroller (OpenMV H7 R2).