I have a pre-trained keras CNN model but not the code used to initially construct the model. I am able to do model.summary() and see the model architecture clearly.
However, in many of the conv layers, very specific parameters were used. For example, certain layers had different dilation rates, strides, initializations, etc.
Is there a way to reverse engineer the model construction code given the trained model? I've looked everywhere for this,but can't seem to find the right wording to search for it.
I can decode a lot of the parameters by manually sifting through the layers one-by-one. But I can't for some of the info like the initializations. Any help would be great.