Search window size in calcOpticalFlowPyrLK is interpreted differently than in cvCalcOpticalFlowPyrLK (Bug #344)

I noticed some performance differences too, but I cannot say it is because of the window size.

I ported my code into full C++ but then I had to switch it to C for the optical flow part because of the different results: but to me also doubling the optical flow size in C++ brings different results than C. Much of the C code was adapted by the optical flow example under samples/c. It would be nice to translate it in C++.

Here is an example:

vector<CvPoint2D32f> _frameAFeatures;
for (int i=0;i<maxCorners;i++) {
_frameAFeatures.push_back(cvPoint2D32f(frameAFeatures[i].x,frameAFeatures[i].y));

}

vector<CvPoint2D32f> _frameBFeatures=vector<CvPoint2D32f>(maxCorners,cvPoint2D32f(0,0));

CvPoint2D32f *fAfts=&_frameAFeaturesr0;
CvPoint2D32f *fBfts=&_frameBFeaturesr0;
char *offts=(char *)&oFFoundFeaturesr0;
float *errs = (float *)&opticalFlowFeatureErrorr0;

/* This is the window size to use to avoid the aperture problem (see slide "Optical Flow: Overview"). */
CvSize optical_flow_window = cvSize(3,3);

/* This termination criteria tells the algorithm to stop when it has either done 20 iterations or when

• epsilon is better than .3. You can play with these parameters for speed vs. accuracy but these values
• work pretty well in many situations.
  */
  CvTermCriteria optical_flow_termination_criteria = cvTermCriteria( CV_TERMCRIT_ITER | CV_TERMCRIT_EPS, 20, .3 );

  IplImage * pyramid1=cvCreateImage(cvSize(frame.cols,frame.rows),IPL_DEPTH_8U,1);
  IplImage * pyramid2=cvCreateImage(cvSize(frame.cols,frame.rows),IPL_DEPTH_8U,1);
  cvCalcOpticalFlowPyrLK(&imgAmt, &imgBmt, pyramid1, pyramid2, fAfts, fBfts, maxCorners, optical_flow_window, 5, offts, errs, optical_flow_termination_criteria, 0 );
  //cvCalcOpticalFlowPyrLK(imgA, imgB, pyramid1, pyramid2, frameAFeatures, frameBFeatures, numOfFeatures, optical_flow_window, 5, oFFoundFeatures, optical_flow_feature_error, optical_flow_termination_criteria, 0 );

  for (int i=0;i<maxCorners;i++) {
  frameBFeatures.push_back(Point2f(_frameBFeatures[i]));
  }

#else
Size _opticalFlowWindw=Size(3,3);
TermCriteria _opticalFlowTerminationCriteria=TermCriteria( CV_TERMCRIT_ITER | CV_TERMCRIT_EPS, 20, .3 );
calcOpticalFlowPyrLK(imgA,imgB,frameAFeatures,frameBFeatures,oFFoundFeatures,opticalFlowFeatureError,_opticalFlowWindw,5,_opticalFlowTerminationCriteria,0.5,0);
#endif

I use self compiled OpenCV 2.1, Linux Ubuntu 9.10, g++ 4.4.1

I also can confirm in quality and performance regression in c++ code