Merge architecture


#1

I want to create a merge architecture with Gluon and fail to find a working example. Essentially I want to have a CNN based architecture receiving input on one side, and a LSTM based architecture on the other side. Than, I want to merge their output and use it as input to another series of layers. Can anyone provide an example?


#2

In the visual question answering example, one part of the model (CNN) takes an image as input, another part of the mode (RNN) takes a question as the input, output of both models are merged and the merged data is provided to another model which is trained to predict answers. Does this help?


#3

Yeah, seems like it is what I want. Thanks for pointing it out.