Coordinating smaller python tflite examples as direct dropins for higher end microcontrollers

[Michael O'Cleirigh]

Hello, 

I´ḿ going to be working on taking smaller tflite python examples to make them work directly and with minimal changes in my tensorflow for micropython firmware.

https://github.com/mocleiri/tensorflow-micropython-examples

As part of this I will be adjusting my micropython facade of the c++ tflm api to be a proper subset of the tflite python api: https://github.com/mocleiri/tensorflow-micropython-examples/issues/10

Mostly I am working on example boards with higher end specs including fast processors, dsp instructions and psram support.

For example:

• Nucleo h743zi2 board (cortex m7 at 480 Mhz with 1MB ram) 
• Teensy 4.1 (cortex m7 at 600 Mhz with support for 16 MB PSRAM).  
• OpenMV has the https://openmv.io/products/openmv-cam-h7-plus which is the h743zi2 but with 32MB psram.
• ESP32 (dual core Tenslica LX6 at 240 Mhz) can support 4MB psram 
• ESP32S3 (dual core Tenslica LX7 at 240 Mhz) can support 8MB and maybe more off chip.

For example micro_speech can run in micropython with live data acquisition via I2S on an esp32 with no spram.  https://github.com/mocleiri/tensorflow-micropython-examples/tree/main/examples/micro-speech/esp32

I think it would be interesting to run a more complex audio model or image processing models to push the capacities of these larger microcontrollers.

Regards, 

Michael