search

henrygouk / nnet

High performance Artificial Neural Network library and utilities
BSD 3-Clause "New" or "Revised" License
1 stars 1 forks source link

Add zero-pad options for Convolutional layer #4

Open dazhenseu opened 8 years ago

dazhenseu commented 8 years ago

Hi~ It seems that your Convolutional layer don't have zero-pad options for input feature map,for example, pad 28x28 to 32x32 with zeros. And it may not be conventient for CNN, so I tried to do some fixes based on your codes. ConvolutionalPadded . But it seems could not run correctly, would you give some advice?:)

henrygouk commented 8 years ago

The easiest way to do this would be to create a Pad layer that only does zero-padding, then add one of these immediately before the Convolutional layer you want to use padding.

If I get some time over the next day or two I will do this, or if you have time you can do it and submit a pull request.