tensorflow / tflite-micro

Infrastructure to enable deployment of ML models to low-power resource-constrained embedded targets (including microcontrollers and digital signal processors).
Apache License 2.0
1.93k stars 827 forks source link

[REQUEST] Please add port TensorFlow Lite Micro to the new RP2350 (and old RP2040) microcontroller platforms (e.g. new Raspberry Pi Pico 2 and the original Raspberry Pi Pico + third-party "Picos")? #2730

Open Hedda opened 4 weeks ago

Hedda commented 4 weeks ago

Can you please add official port of the TensorFlow Lite Micro library to new RP2350 (and old RP2040) microcontroller platforms?

There looks to be a few unofficial ports of the TensorFlow Lite Micro library for the RP2040 but obviously RP2350 should perform better.

There is LiteRT downstream port for older RP2040 but does't look to be maintained + has not been updated to new RP2350, see:

and

RP2350 is the successor and next-generation of RPi RP2040 so guess ESPHome would need to update the RP2040 Platform for it or add another platform?

and/or

RP2350 is newly released so there is not yet widespread support but RP2040 based development boards are already very popular in schools and STEM education so think they are interesting as alternatives to ESP32 because documentation is often easier as often targeting beginners which in turn means can lead to broader use and use cases.

FYI, there is not yet an official Raspberry Pi Pico 2 board with Wi-Fi but there are third-party boards with WiFi and LAN, see:

Hedda commented 4 weeks ago

If added it by the way would be very cool if could get Raspberry Pi to then make a demo video showing microWakeWord running natively on RP2350 or RP2040 microcontrollers and maybe post a blog post about it a show case for tflite-micro and Raspberry Pi Pico 2!

microWakeWord (also known as "mWW") is a fully open-source wake word detection library made for MCU-based voice assistants, and it depends on the TensorFlow Lite (LiteRT) library to run fast nativley on low-power microcontrollers:

The microWakeWord was created by Kevin Ahrendt and it enables for example the ESPHome firmware (via microphones) to on-device wake words detections on microcontroller devices.

If you do please post a new blog post to announce the updated and added RP2350 support!

PS: The microWakeWord project as been adopted/transfered to the Open Home Foundation so it will always remain as open-source:

github-actions[bot] commented 4 days ago

"This issue is being marked as stale due to inactivity. Remove label or comment to prevent closure in 5 days."

Hedda commented 4 days ago

Still wanted.