Possible native stack corruption in C++ API's findContours(Mat, Mat, Mat, int, int, Point)

[chile12]

Hello

I encountered a SIGSEGV when executing my javacv based heuristic in a loop.
After a lot of searching, reading and experimenting I found the source has to be the findContoures method (org.bytedeco.javacpp.opencv_imgproc.findContours).
The only workaround I came up with was using the C implementation of this method.
SIGSEGV disappeared instantly.

Let me know if you need my code example for further investigation.

Greetings.

PS.: I#m using v. 0.9

[saudet]

Could you provide a few lines of code to reproduce that issue? Thanks!

[chile12]

All right:

This is a short extract of my code which is executed constantly in a loop.
In the second loop, the first allocation of an opencv object raises the SIGSEGV:

08-28 18:19:25.671: I/DEBUG(3741): signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr deadbaad

public static  List<Rect> getBoundingBoxesOfContoures(Mat bitImage, BackGroundShade shade, int blockSize, int weightConstant)  throws FrameNotFoundException
{
    Mat contourImage = bitImage;
    List<Rect> boundingBoxes = new ArrayList<Rect>();
    MatVector contours = new MatVector();
    if(shade == BackGroundShade.Dark || shade == BackGroundShade.Undef)
    {
        adaptiveThreshold(bitImage, contourImage, 255, CV_ADAPTIVE_THRESH_GAUSSIAN_C, CV_THRESH_BINARY, blockSize, weightConstant);
        contours.put(ImageStatics.GetContoursOfImage(contourImage));
        boundingBoxes.addAll(extractRectanglesFromMatVector(contours));
    }
    if(shade == BackGroundShade.Bright || shade == BackGroundShade.Undef)
    {
        adaptiveThreshold(bitImage, contourImage, 255, CV_ADAPTIVE_THRESH_GAUSSIAN_C, CV_THRESH_BINARY_INV, blockSize, weightConstant);
        contours.put(ImageStatics.GetContoursOfImage(contourImage));
        boundingBoxes.addAll(extractRectanglesFromMatVector(contours));
    }
    return boundingBoxes;
}
public static MatVector GetContoursOfImage(Mat input)
{
    MatVector contours = new MatVector();
    findContours(input, contours, new Mat(), CV_RETR_CCOMP, CV_CHAIN_APPROX_SIMPLE, new Point(0,0));

    return contours;
}

public static List<Rect> extractRectanglesFromMatVector(MatVector contours) throws FrameNotFoundException {
    List<Rect> outout = new ArrayList<Rect>();
    for (int i =0; i < contours.size(); i++)
    {
        Rect zw = boundingRect(contours.get(i));
        if(zw != null)
            outout.add(zw);
    }
    return outout;
}

My current work araound looks like this:

public static List<CvRect> GetContoursOfImage(Mat input)
{
    CvSeq contours = new CvSeq();

    cvFindContours(input.asCvMat(), CvMemStorage.create(), contours, Loader.sizeof(CvContour.class), CV_RETR_LIST, CV_CHAIN_APPROX_SIMPLE);

    List<CvRect> rects = new ArrayList<CvRect>();
    CvSeq zw = new CvSeq();
    if(contours!=null) {
        for (zw = contours; zw != null; zw = zw.h_next()) { 
            rects.add(cvBoundingRect(zw,0));
        }

    }   
    return rects;
}

I'd like to see this without the C-work-around :)

Thank you.

[saudet]

Your issue might have something to do with your put() call. There isn't a MatVector.put(MatVector) method, so what gets called is MatVector.put(Pointer) and that uses memcpy(), which is probably not a good idea for a non-PODS class...

[chile12]

Seems kinda obvious in hindsight, damn it.

Thanks.