TensorFlow Lite Micro support for machine learning

[NanWang]

Hello,

Up to now, is there official Nordic support on Tensorflow Lite Micro for nRF52840?

Thanks.

BRs.

[JONATHAN LL]

Hi,

The use of Edge impulse is what you could consider officially supported machine learning. 

 Using Edge Impulse with the nRF Connect SDK — nRF Connect SDK 1.6.99 documentation (nordicsemi.com)

What is embedded ML, anyway? (edgeimpulse.com)


Here is a example on the nrf5240 DK https://docs.edgeimpulse.com/docs/nordic-semi-nrf52840-dk 


Here is a blog post on TF (+) TensorFlow and nRF - Blogs - Nordic Blog - Nordic DevZone (nordicsemi.com)

Regards,
Jonathan

[NanWang]

Hi,

Thank you for your information. I have already known what you provide and used Arduino Nano 33 BLE Sense through Arduino IDE and Edge Impulse which already support TF Lite Micro(examples). Now we want to move nRF52840 but we can't find Nordic official implementation of TF Lite Micro on nRF52840. So we don't know what how to support TFLM on nRF52480 without using third-party platforms. Any other good suggestions?

Thank you.

BRs

[JONATHAN LL]

Hi,

I am sorry but we do not have a official library support for TF Lite Microcontroller. Following the Blog is the closest you can get if you don't want to use third party. I think that Edge impulse is the quickest options, then you also have the Edge impulse forum to ask for help.

There is the zephyr sample as well, sdk-zephyr/samples/modules/tensorflow at master · nrfconnect/sdk-zephyr (github.com). 

I also did some testing 1+ year back and got some samples working nicknameBOB/TF_NCS_dev: Tensorflow examples in zephyr on nrf products (github.com), it is the same method as in the Devzone blog.

Regards,
Jonathan