Streaming ML models through NPU on i.MX93 Yocto V6.6.3

[mohamed_gaseen]

Hi,

I'm currently working on streaming the machine learning models such as face recognition, DMS and more on my i.MX93 v6.6.3 through NPU instead of CPU.

The face recognition become a success for me, I'll share you the procedure through text.

But the thing is the driver monitoring system(dms) is not working not even through CPU.

As the issue is with the download_models.py links are invalid

face_detection_front_128_full_integer_quant.tflite: HTML document, ASCII text, with very long lines (35986)

Is there any platform to generate the ML models as for the CPU and NPU quantization.

Is there any possible way to generate the models and get the valid ".tflite" files.

Please share me a solution ASAP. As this is an emergency requirement.

[Chavira]

Hi @mohamed_gaseen!

Thank you for contacting NXP Support!

You can see the link taht we use to do download the models directly in our Github.

For some reason google and other companies change the link where we can download the models and the script download_models.py is not updated.

If you have problems running the DMS example please let me know.

Best Regards!

Chavira

[mohamed_gaseen]

Thank you for the timely reply @Chavira 

I've been searching in a lot of sources but can't find a valid source or any way for running the Machine Learning DMS model through CPU or NPU as I expected.

As my requirement is completely into the automation now, The customer wants to stream DMS through NPU, that's the major work to do!

Please kindly share me a valid way to download or generate the "tflite" source into a vela model.

Thanks in advance.

[Chavira]

Hi @mohamed_gaseen!

It is easy!

You can use eIQ tools to convert the model or use the command vela in the imx93 board.

You can see more about that in our Machine Learning User Guide

[mohamed_gaseen]

Hi @Chavira 

All the three files are invalid/corrupted, Please provide me a valid link of those below required ".tflite" files.

root@imx93EVK:/usr/bin/eiq-examples-git/dms# ls ../models/face_detection_front_128_full_integer_quant.tflite
../models/face_detection_front_128_full_integer_quant.tflite

root@imx93EVK:/usr/bin/eiq-examples-git/dms# ls ../models/face_landmark_192_integer_quant.tflite
../models/face_landmark_192_integer_quant.tflite

root@imx93EVK:/usr/bin/eiq-examples-git/dms# ls ../models/iris_landmark_quant.tflite
../models/iris_landmark_quant.tflite

Thanks in Advance.

[mohamed_gaseen]

Helllo @Chavira 

Followed the Machine learning document, mentioned nothing about the Driver Monitoring System(DMS).

Is there any specific document or source about the DMS. If so it would be very helpful.

The "tflite" file required for the DMS is invalid from the download_models.py link, that's the main reason for this fault.

The below attached log is for your reference:

root@imx93EVK:/usr/bin/eiq-examples-git/models# file face_detection_front_128_full_integer_quant.tflite
face_detection_front_128_full_integer_quant.tflite: HTML document, ASCII text, with very long lines (31211)

Please provide me the valid DMS model running file, As the face recognition works successfully. Needs to stream the Driver Monitoring System through  NPU too.

Thanks in Advance.

[Chavira]

HI @mohamed_gaseen!

To download the proper models you should run the python script called "download_models.py", but I recommend to download the latest script available.

Best Regards

Chavira