How i can build the .model file for APEX DNN?

[woong214]

Hello

How i can build the .model file for APEX DNN?

I want to use my own Neural Network at S32v234evk2.

I installed S32v VSDK and saw the CNN_classifier example.

in example and APEX DNN pdf file, squeezeNet's weight is loaded in squeezeNet.model.

I don't know how to build this weight file. I want to make the basic MLP.model

In attached pdf file, this model is trained by Caffe.

After training, What should I do? Quantize Caffe model & Fine-tune quantized model & Test quantized model...
How Can I do this step?

or Is there any other way to use a neural network?

[paulvlase]

Hi woong214@postech.ac.kr‌,

The existing DNN example from the Vision SDK implements only the Squeezenet architecture, it doesn't work with custom ones.

You can use the Matlab Deep Learning Toolbox to run your custom network on ARM. Our toolbox has a nxpvt.CNN wrapper to generate the code for the network and run in on the board. We have AI and Machine Learning Course that shows how to do that.