Possible to convert CDCN network under DSDG to onnx?

[kadirbeytorun]

Hello,

While trying to convert the CDCN network to onnx, network gets messed up for some reason and cannot be used for inference later. Can I use the model convertor examples under addition_module for this task? Could you guide me on what changes I should make to the convert scripts for this?

Furthermore, I am quite confused about the output of this network. It seems output's shape is 32x32. How should I evaluate this result? Is it like, if I take their average for "real" classes average will be higher? Couldn't understand this part from your article.

Regards, Kadir