Exporting Matlab trained CNN model to ONNX Problem

Hello,

I exported CNN .mat file to .onnx using exportONNXNetwork function so I can use it with tensorflow-lite for android use.

The ONNX model now has an input size shape of 3 x 224 x 224 (when I visualize the model using Netron - image attached) instead of 224 x 224 x 3 (Matlab model architecture) :

error : The input tensor should have dimensions 1 x height x width x 3. Got 1 x 3 x 224 x 224.

Is there any solution to fix this problem so i can use the CNN onnx model for image classification prediction in tflite?

p.s: I think that all the layers dimensionality are inversed not only the input layer shape.

Thank you!

Answer 1

[Sutanu Maiti]

MATLAB trained CNN models and be exported and saved as a .mat and can directly be used for deploymnet on Android device by using "Simulink support package for Android devices".

You may have a look at the example below:

https://www.mathworks.com/help/supportpkg/android/ref/detect-and-track-objects-using-deep-learning-on-android-device.html