libeblearn
|
#include <ebl_basic.h>
Public Member Functions | |
mirrorpad_module (int nr, int nc) | |
mirrorpad_module (idxdim &kernel_size) | |
virtual | ~mirrorpad_module () |
destructor | |
virtual void | fprop (Tstate &in, Tstate &out) |
forward propagation from in to out | |
virtual void | fprop (Tstate &in, idx< T > &out) |
forward propagation from in to out | |
virtual mirrorpad_module< T, Tstate > * | copy (parameter< T, Tstate > *p=NULL) |
A simple mirror padding module that is mostly usefull for doing same size output convolutions.
ebl::mirrorpad_module< T, Tstate >::mirrorpad_module | ( | int | nr, |
int | nc | ||
) |
<nrow> is the number of rows in zero added border <ncol> is the number of cols in zero added border the output size is enlarged by 2*nrow in rows and 2*ncols in cols for each feature map.
ebl::mirrorpad_module< T, Tstate >::mirrorpad_module | ( | idxdim & | kernel_size | ) |
Constructor adding zero borders with same size on each size if the kernel had odd size, otherwise adding 1 pixel less on the right and bottom borders.
kernel_size | The sizes of the kernel for which to pad. |
mirrorpad_module< T, Tstate > * ebl::mirrorpad_module< T, Tstate >::copy | ( | parameter< T, Tstate > * | p = NULL | ) | [virtual] |
Returns a deep copy of this module.
p | If NULL, reuse current parameter space, otherwise allocate new weights on parameter 'p'. |
Reimplemented from ebl::zpad_module< T, Tstate >.