MATLAB Answers

What is corresponding implementation in Keras layers for groupedCon​volution2d​Layer ?

4 views (last 30 days)
Jhih-Siang Lai
Jhih-Siang Lai on 22 Mar 2021
Answered: Prateek Rai on 14 Sep 2021
I am implementing layers for both Keras and MATLAB, but not sure how MATLAB implements groupedConvolution2dLayer.
Ahough MATLAB document describes groupedConvolution2dLayer could be converted into either DepthwiseConv2D or SeparableConv2D,
It seems like groupedConvolution2dLayer did more than that.
Does anyone know how to convert this into Keras layer or Tensorflow?
filterSize = [7 7];
numFiltersPerGroup = 16;
numGroups = 4;
groupedConvolution2dLayer(filterSize,numFilters,numGroups,"DilationFactor",1,"Name","groupedconv1","Padding","same")
Many thanks

Answers (1)

Prateek Rai
Prateek Rai on 14 Sep 2021
To my understanding, you wanted to implement groupedConvolution2dLayer into Keras/Tensorflow.
One possible workaround is you can use 'exportONNXNetwork' to export whole network to ONNX model format. Then, you can use the ONNX Model to import it to Keras/Tensorflow.
You can refer to exportONNXNetwork MathWorks documentation page to find more on exporting network to ONNX model format.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!