Closed ghost closed 8 years ago
@seyounglee95 I'll do a pass
This close #172
im2col
and im2row
is almost same except the shape of an output matrix and the indices of its elements. How about having another kind of inner function that has an argument indicating row form or column form? im2col
and im2col
can call this function with the argument.
It will be better to remove im2col
and col2im
since we don't use these methods now.
If these are needed later, we can easily implement these as you did.
I remember that I recommended you to leave these methods. Sorry for confusing you.
@seyounglee95 This looks awesome! I'm merging this.
This PR includes support for 3D input data and multi-filters.
num_output
decides the number of the filters. As new parameter is added in convolutional layer, configuration parts related to that layer are modified. For convenience of computation,im2row()
androw2im()
are implemented. A few test cases with 3D input data are added, too.