Is it possible to define a more complex network?

[Rice2999]

Hi, Hagaygarty

Sorry for so many questions. (Because I love your work and am currently having fun with it.)

It seems that, with mdCNN, it is easy to define standard CNN network structures, like this one: https://cdn-images-1.medium.com/max/1600/1*XbuW8WuRrAY5pC4t-9DZAQ.jpeg

Is it possible to define a network with branches? Like this http://www.mdpi.com/applsci/applsci-07-00447/article_deploy/html/images/applsci-07-00447-g002.png

With up-pooling? Like this https://leonardoaraujosantos.gitbooks.io/artificial-inteligence/content/image_folder_7/Deconvnet.png

And with multiple separated CNN network and a fusion step? Like this https://ai2-s2-public.s3.amazonaws.com/figures/2017-08-08/465dedb91349eda9a33b3b3b96ff3036b337695e/2-Figure1-1.png

Thanks.

-Deshan

[Hagay Garty]

Hi, you cannot do it natively, but for your first example, you can simply run 2 networks in parallel and then combine both outputs softmax formula. The networks are trained separately

As for the second example, its cannot be done without coding. much coding actually :)