[mlpack-git] [mlpack/mlpack] Inception layer (#757)

[Marcus Edel]

> +    // Forward pass for 3x3 pool path. 
> +    pool3.InputParameter() = input;
> +    pool3.Forward(input, pool3.OutputParameter());
> +    Pad(pool3.OutputParameter(), 1, 1, convPool.InputParameter());
> +    convPool.Forward(pool3.OutputParameter(), convPool.OutputParameter());
> +    biasPool.Forward(convPool.OutputParameter(), biasPool.OutputParameter());
> +    basePool.InputParameter() = biasPool.OutputParameter();
> +    basePool.Forward(convPool.OutputParameter(), basePool.OutputParameter());
> +
> +    // concatenate outputs of all the paths.
> +    output = arma::join_slices( 
> +              arma::join_slices(
> +                arma::join_slices( 
> +                  base1.OutputParameter(), base3.OutputParameter() ), 
> +                  base5.OutputParameter() ), basePool.OutputParameter());
> +

As you already pointed out the inception layer is basically a container for a couple of convolution and pooling layer. Since they are abstracted through the inception layer, the main network has now knowledge about the containing layer. So, the inception layer has to forward any weights to any capsulated layer.


-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/mlpack/mlpack/pull/757/files/06d923321f246f2c6ead9ad56e99309fe24a6f5c#r75576884
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.cc.gatech.edu/pipermail/mlpack-git/attachments/20160820/90b8cae1/attachment.html>