PatchExtractor Class Reference

Class for extracting spatio-temporal patches (cubes) from a video (online mode). More...

#include <PatchExtractor.hpp>

Collaboration diagram for PatchExtractor:
[legend]

List of all members.

Public Types

enum  fill_type { FILL_NN, FILL_NNFILL, FILL_VOTE }
 

fill type for generating images from patch labels.

More...

Public Member Functions

 PatchExtractor (const PatchOptions &patopt, int vrows, int vcols)
 create a new PatchExtractor.
void reset ()
 reset the buffer.
int framesToNextPatch ()
 returns the number of frames needed to complete next patch.
bool addFrames (const Mat &vid)
 add frames to the buffer.
const vector< Mat > & getPatches ()
 return the last formed patches.
const Mat & getCenterFrame ()
 return the center frame of the video buffer.
const Mat & getPatchFrames ()
 return the frames (Mat video) associated with the current set of patches.
const Mat & getbuf ()
 return the current video buffer.
const Mat & getbuf_zm ()
 return the current zero-mean video buffer.
void patchLabelsToImage (const Mat &patLabels, Mat &imgLabels, fill_type fillMode=FILL_NN)
 convert patch labels to an image.
void dump (ostream &output) const
 debugging: dump to an output stream.
void dump_matlab_init (ostream &output) const
 debugging: dump initial info to a matlab-readable output stream (pat struct).
void dump_matlab_patches (ostream &output) const
 debugging: dump current patches to a matlab-readable output stream (pat struct).

Public Attributes

PatchOptions patopt
 options for extracting patches.
int vrows
 number of rows in the video frame.
int vcols
 number of columns in the video frame.
vector< Point2i > locyx
 (y,x) coordinates of top-left of each patch.
vector< bool > locyx_mask
 patch mask depending on pixel variance.
int locz
 current z location of the patches (z of the first frame of the patch).
int curz
 z location for next frame added.
int nextz
 value of curz that will form a new patch.
Point3i coff
 offset to center of patch.
vector< int > allx
 all x-locations on step grid.
vector< int > ally
 all y-locations on step grid.
Mat vbuf
 video buffer to store current frames [vrows,vcols,patopt.win.z] type=OPT_MAT_TYPE.
Mat vbuf_zm
 video buffer w/o mean [vrows,vcols,patopt.win.z] type=OPT_MAT_TYPE.
vector< Mat > patches
 the set of patches.
bool flag_zm
 flag if vbuf_zm is used.
Mat frame_center
 center frame of the video buffer (reference to vbuf).
Mat frame_patches
 center frames of the video buffer (reference to vbuf).

Private Member Functions

void getLoc ()
 get all locations that might have a patch.
void processNextPatch ()
 process patches after they have been formed.

Friends

ostream & operator<< (ostream &output, const PatchExtractor &cov)
 debugging: dump to an output stream.

Detailed Description

Class for extracting spatio-temporal patches (cubes) from a video (online mode).

The extractor works in an "online" mode, where frames are added sequentially. When enough frames are added a true flag is returned, which indicates that new patches have been formed. Normalization of patches is handled efficiently, and access is given to the current video buffer. Also included is a function for converting a set of labels corresponding to each patche to an image.

The actual patches will be in different places, depending on the normalization used: 1) norm_none --> patches are sub-videos of the buffer (vbuf); 2) norm_zm --> patches are sub-videos of the buffer (vbuf_zm); 3) norm_zmuv --> patches are local copies. Note that in some cases, the patches are just references to a sub-videos of the video buffer. Hence, it is not allowed to modify the patches.

See also:
PatchOptions, PatchBatchExtractor

Definition at line 34 of file PatchExtractor.hpp.


Member Enumeration Documentation

fill type for generating images from patch labels.

See also:
patchLabelsToImage
Enumerator:
FILL_NN 

use nearest-neighbor.

FILL_NNFILL 

use nearest-neighbor (fill in patches).

FILL_VOTE 

overlapping patches vote.

Definition at line 41 of file PatchExtractor.hpp.


Constructor & Destructor Documentation

PatchExtractor::PatchExtractor ( const PatchOptions &  patopt,
int  vrows,
int  vcols 
)

create a new PatchExtractor.

Parameters:
patopt patch options.
vrows number of rows in the video frame.
vcols number of columns in the video frame.

Definition at line 38 of file PatchExtractor.cpp.

Here is the call graph for this function:

Here is the caller graph for this function:


Member Function Documentation

bool PatchExtractor::addFrames ( const Mat &  vid  ) 

add frames to the buffer.

Parameters:
vid the Mat video to add (all frames are added).
Returns:
true if new patches have been formed, false otherwise.

Definition at line 177 of file PatchExtractor.cpp.

Here is the call graph for this function:

Here is the caller graph for this function:

void PatchExtractor::dump ( ostream &  output  )  const

debugging: dump to an output stream.

Definition at line 485 of file PatchExtractor.cpp.

void PatchExtractor::dump_matlab_init ( ostream &  output  )  const

debugging: dump initial info to a matlab-readable output stream (pat struct).

Definition at line 497 of file PatchExtractor.cpp.

void PatchExtractor::dump_matlab_patches ( ostream &  output  )  const

debugging: dump current patches to a matlab-readable output stream (pat struct).

Definition at line 509 of file PatchExtractor.cpp.

Here is the call graph for this function:

int PatchExtractor::framesToNextPatch (  ) 

returns the number of frames needed to complete next patch.

Definition at line 172 of file PatchExtractor.cpp.

const Mat & PatchExtractor::getbuf (  ) 

return the current video buffer.

Definition at line 391 of file PatchExtractor.cpp.

const Mat & PatchExtractor::getbuf_zm (  ) 

return the current zero-mean video buffer.

Only valid if zero-mean normalization is used.

Definition at line 395 of file PatchExtractor.cpp.

const Mat & PatchExtractor::getPatchFrames (  ) 

return the frames (Mat video) associated with the current set of patches.

The number of frames will be the z-step size. TODO: Except on the first call, which will also include the beginning frames.

Definition at line 387 of file PatchExtractor.cpp.

void PatchExtractor::patchLabelsToImage ( const Mat &  patLabels,
Mat &  imgLabels,
fill_type  fillMode = FILL_NN 
)

convert patch labels to an image.

Works when patLabels has channels, which are passed to imgLabels. (implementation of segm_mask.m)

Parameters:
patLabels the label for each patch.
imgLabels output image (has the same type as patLabels).
fillMode mode for filling in patches.

Definition at line 403 of file PatchExtractor.cpp.

void PatchExtractor::processNextPatch (  )  [private]

process patches after they have been formed.

applies the normalization.

Definition at line 265 of file PatchExtractor.cpp.

Here is the call graph for this function:

Here is the caller graph for this function:

void PatchExtractor::reset (  ) 

reset the buffer.

i.e. start a new video.

Definition at line 163 of file PatchExtractor.cpp.

Here is the caller graph for this function:


Friends And Related Function Documentation

ostream& operator<< ( ostream &  output,
const PatchExtractor &  cov 
) [friend]

debugging: dump to an output stream.

See also:
dump

Member Data Documentation

vector<int> PatchExtractor::allx

all x-locations on step grid.

Definition at line 55 of file PatchExtractor.hpp.

vector<int> PatchExtractor::ally

all y-locations on step grid.

Definition at line 56 of file PatchExtractor.hpp.

offset to center of patch.

Definition at line 54 of file PatchExtractor.hpp.

z location for next frame added.

Definition at line 52 of file PatchExtractor.hpp.

flag if vbuf_zm is used.

Definition at line 62 of file PatchExtractor.hpp.

center frame of the video buffer (reference to vbuf).

Definition at line 63 of file PatchExtractor.hpp.

center frames of the video buffer (reference to vbuf).

Definition at line 64 of file PatchExtractor.hpp.

vector<Point2i> PatchExtractor::locyx

(y,x) coordinates of top-left of each patch.

Definition at line 49 of file PatchExtractor.hpp.

patch mask depending on pixel variance.

Definition at line 50 of file PatchExtractor.hpp.

current z location of the patches (z of the first frame of the patch).

Definition at line 51 of file PatchExtractor.hpp.

value of curz that will form a new patch.

Definition at line 53 of file PatchExtractor.hpp.

the set of patches.

Definition at line 61 of file PatchExtractor.hpp.

options for extracting patches.

Definition at line 46 of file PatchExtractor.hpp.

video buffer to store current frames [vrows,vcols,patopt.win.z] type=OPT_MAT_TYPE.

Definition at line 59 of file PatchExtractor.hpp.

video buffer w/o mean [vrows,vcols,patopt.win.z] type=OPT_MAT_TYPE.

Definition at line 60 of file PatchExtractor.hpp.

number of columns in the video frame.

Definition at line 47 of file PatchExtractor.hpp.

number of rows in the video frame.

Definition at line 47 of file PatchExtractor.hpp.


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

Generated on 18 Dec 2014 for libdt-OpenCV by  doxygen 1.6.1