libeblearn
ebl::channorm_module< T, Tstate > Class Template Reference

Abstract class for normalization of image channels. More...

#include <ebl_preprocessing.h>

Inheritance diagram for ebl::channorm_module< T, Tstate >:
ebl::channels_module< T, Tstate > ebl::module_1_1< T, Tstate > ebl::module ebl::bgr_to_yp_module< T, Tstate > ebl::bgr_to_ypuv_module< T, Tstate > ebl::rgb_to_hp_module< T, Tstate > ebl::rgb_to_rgbn_module< T, Tstate > ebl::rgb_to_yn_module< T, Tstate > ebl::rgb_to_ynunvn_module< T, Tstate > ebl::rgb_to_ynuv_module< T, Tstate > ebl::rgb_to_ynuvn_module< T, Tstate > ebl::rgb_to_yuvn_module< T, Tstate > ebl::y_to_yp_module< T, Tstate >

List of all members.

Public Member Functions

 channorm_module (idxdim &kerdim, bool mirror=true, t_norm norm_mode=WSTD_NORM, const char *name="channorm", int nf=1, bool globnorm=true)
virtual ~channorm_module ()
 Destructor.
virtual void fprop (Tstate &in, Tstate &out)=0
 Forward propagation from in to out (abstract).
virtual channorm_module< T,
Tstate > * 
copy ()=0
 Returns a deep copy of this module (abstract).
virtual std::string describe ()
 Returns a string containing describe() string of all modules.
module_1_1< T, Tstate > * new_norm (idxdim &normker, bool mirror, t_norm norm_mode, int nf)
 Returns a new norm module.

Protected Member Functions

virtual void resize_output (Tstate &in, Tstate &out, int dim0=-1)

Protected Attributes

idxdim normker
 norm kernel size
Tstate tmp
 temporary buffer
module_1_1< T, Tstate > * norm
 contrast normalization module
bool mirror
 Mirror padding or not.
t_norm norm_mode
 Normalization mode.

Detailed Description

template<typename T, class Tstate = bbstate_idx<T>>
class ebl::channorm_module< T, Tstate >

Abstract class for normalization of image channels.


Constructor & Destructor Documentation

template<typename T , class Tstate >
ebl::channorm_module< T, Tstate >::channorm_module ( idxdim &  kerdim,
bool  mirror = true,
t_norm  norm_mode = WSTD_NORM,
const char *  name = "channorm",
int  nf = 1,
bool  globnorm = true 
)

Constructor.

Parameters:
kerdimKernel dimensions for local normalization.
mirrorIf true, pad normalization if a mirror of the image instead of with zeros. This can be useful in object detection when objects are close to borders.
norm_modeThe type of normalization (WSTD_NORM by default).
nfThe number of features to normalize across.
globnormNormalize channels globally or not.

Member Function Documentation

template<typename T , class Tstate >
void ebl::channorm_module< T, Tstate >::resize_output ( Tstate &  in,
Tstate &  out,
int  dim0 = -1 
) [protected, virtual]

Resize the output based on input dimensions

Parameters:
dim0An optional size for the first dimension. Set it to 1 when converting from color to greyscale.

Reimplemented from ebl::channels_module< T, Tstate >.


The documentation for this class was generated from the following files: