Question about how to integrate ML model into FRDM-MCXN947 by tflm_ciar10 project

[BillWen]

Hi Everyone:

My environment : 

1. MCUXpressoIDE_11.9.1

2. eIQ Toolkit v1.11.4

3. SDK_2_14_0_FRDM-MCXN947

I try to put my ML model into FRDM-MCXN947 by tflm_ciar10 project, the following is my steps:

1. Use eIQ toolkit to transfer LSTM model from .h5 file to Tensorflow Lite, and transfer again from Tensorflow Lite to Tensorflow Lite for Neutron

2. Follow the document "How to Integrate Customer ML Model to NPU on MCX N94x" -> AN14241 to put my model into project

3. Setup a test data, and put into project. If my test data is .mat or .blf file, can I put inside or I should transfer it to .h file first?

4. build successful

5. power on device and error

CIFAR-10 example using a TensorFlow Lite Micro model.
Detection threshold: 60%
Model: cifarnet_quant_int8_npu
Didn't find op for builtin opcode 'UNIDIRECTIONAL_SEQUENCE_LSTM'
Failed to get registration from op code UNIDIRECTIONAL_SEQUENCE_LSTM

AllocateTensors() failedFailed initializing model

Do anyone have experience on this situation?

BTW, where can I get the demo model and test data according to AN14241 document? I can't find the path or reference about it, thanks.

Thanks a lot

BR, Bill Wen

[Alice_Yang]

Hello @BillWen 

There is simple guide demonstrating the steps. Also includes video. Please refer to it:

https://www.nxpic.org.cn/module/forum/forum.php?mod=viewthread&tid=803297&page=1&extra=#pid3004813  

BR

Alice

[BillWen]

Hello Alice:

First of all. Thanks a lot for your help.

I still don't know why "Label CIFAR10 image programs" without example model for test. I follow the AN14241 application note mentions the converted file "custom_model_converted.tflite", but converted file not provide in SDK.

How to get converted example file or RAW file(before converted)?