My question is as follows: I’m building a binary classification model that combines 2 types of input (images and associated numerical data). There are many examples of such mixed input models: a few layers for type 1 input processing, a few layers for type 2 input processing, then both concatenate and final layers are applied to the concatenated input.
However, I’ve stumbled across a problem with how to incorporate transfer learning in such a model.
i.e. I would like to use Resnet to classify the images instead of some custom layers. Is there any way to do it?