Question of Tensor flow Lite

[Seongyon_Jeong]

I am user of I.MX RT1020, and also,

I am interested in developing "ON Device AI" by porting AI libraries to embedded devices.

To be honest, I don't have any basic knowledge of AI libraries. So, when I checked, I found that there are solutions like TensorFlow Lite, PyTorch Mobile, ONNX Runtime, ARM CMSIS-NN ... etc.

But, the SDK for the I.MX RT1020-EVK I'm using, doesn't provide these stacks from NXP.

Do I need to upgrade the RT1020 platform a little more? For example, RT1080? Or S32G with a completely different chip? Different NXP products?

If that's not the case,
Do I have to port the tensor flow Lite on RT1020-EVK board, myself?

I think there was a similar question in 2017 year (https://community.nxp.com/t5/i-MX-Processors/How-to-add-tensor-flow-library-to-Yocto-layer/m-p/67470...) , Is not there yet a solution or alternative to the AI framework or library in NXP company?

Please share the data or contents.

[Gavin_Jia · Accepted solution]

Hi @Seongyon_Jeong ,

Thanks for your interest in NXP MIMXRT series!

Part of the RT series is the provision of machine learning capabilities. When building the SDK, if eIQ support is provided, then the corresponding machine learning demos will be available in the SDK.

And our engineers have provided some other full example projects that can help you get started quickly, hopefully these resources below are what you're looking for:
1. Handwritten Digit Recognition Using TensorFlow Lite Micro on i.MX RT devices
2. eIQ Getting Started with i.MX RT
3. The “Hello World” of TensorFlow Lite
4. eIQ® Transfer Learning Lab with TensorFlow Lite for i.MX RT
5. Getting Started with TensorFlow Lite for Microcontrollers on i.MX RT
6. https://www.nxp.com/design/training/develop-ml-applications-with-the-glow-neural-network-compiler-an...
7. https://www.nxp.com/design/training/machine-learning-with-the-i-mx-rt1060-crossover-mcu:TIP-MACHINE-...

Best regards,
Gavin

[Seongyon_Jeong]

That's wonderful
Thank you for linking a lot of data.
First of all, I'll have to analyze the data you linked.
Thank you for your answer, I will post the question again if necessary.