How to start with Model training and vision based application in MCXN947

Gayathriii

I ran the face detect ACH code, it's working. I want to know more about the BYOM and training a model from scratch in the eIQ toolkit. In the last conference, NXP mentioned something about BYOM and BYOD.

I need to upload my images, label it and train and want to put it in MCXN947 board. Is it possible?

I tried accessing the eIQ model creator, the request has been submitted. I have downloaded the eIQ timeseries (I think it's for data like motor anomaly detection or something) and also downloaded the eIQ neutron SDK. I just want the workflow or material or user guide for training the model and put it in MCXN947.

Gayathriii_0-1788414712451.png

Harry_Zhang

Hi @Gayathriii

Yes, this is absolutely possible on the MCXN947.

For your use case:

Upload my images → label them → train a model → run on MCXN947

This is a classic BYOD workflow.

Option 1: eIQ Model Creator

Once your eIQ Model Creator access is approved:

Create a new image classification project.
Upload your image dataset.
Label the images (or organize them by folders/classes).
Train a model using transfer learning.
Export a TensorFlow Lite model (.tflite).
Use the eIQ Neutron SDK to convert the model for MCXN947.
Integrate the converted model into your MCUXpresso project and run it on the board.

Option 2: Train Yourself (TensorFlow)

You do not actually need Model Creator.

You can:

Collect images.
Label them.
Train with TensorFlow/Keras.
Export

Resources：

1. MCXN947: How to Train and Deploy Customer ML Model to NPU

Images
   ↓
Training
   ↓
TensorFlow Lite model
   ↓
neutron-converter
   ↓
NPU-enabled .tflite
   ↓
MCUXpresso project
   ↓
MCXN947 NPU
``

MCXN947：如何训练和部署客户 ML 模型到 NPU

2. AN14241 - How to Integrate Customer ML Model to NPU on MCX N94x

How to Integrate Customer ML Model to NPU on MCX N94x

3. How to Get Started with NPU and ML in MCX Microcontrollers

MCX 微控制器中 NPU 与机器学习的入门指南

BR

Harry

View solution in original post

Harry_Zhang