Are there any plans to add support in exportONNXNetwork tool for the maxUnpooling2dLayer?

Hi,

I have found the maxUnpooling2dLayer to be very useful when working with autoencoder/decoder structures together with deep learning At least for the problem I am working with the alternative to use upsampling via convolutional operations does not seem to work nearly as well as the use of maxunpooling. I now have a working network and would like to the next step and export my network and plug it into an app of my own creation. The MATLAB documentation clearly states that the exportONNXNetwork function does not support the maxUnpooling2dLayer function. It seems however as if the ONNX specification/engine itself does support maxunpooling. I am therefore wondering if there are any plans to add support for the maxUnpooling2dLayer to the exportONNXNetwork tool? If so is it possible to give any guesstimates of when this functionality could be available?

Thanks, Mats

Answer 1

[John D'Errico]

Sorry, but surely not possible to tell you.

MathWorks does not discuss their plans for the future on this forum. Anyone who does know the answer is probably not allowed to disclose the information. They certainly would not tell you the expected release time for something anyway, as if a problem happens to arise, that release date would be wrong.