Nabble - opencvlibrary-devel

Re: Problems with the new svn service...

2009-12-07T23:34:43Z

Hi Vadim,

1. Ok, that's nice - it assures me, that I'm not do dumb to log into an internet account ;) Hope you get this working fast...

2. Yes you're absolutely right, I'm interested in the nightly builds that were provided by your last host (or configuration). In the ViewVC help of your page I found the mentioned option for the tar ball access.
https://code.ros.org/gf/project/opencv/scmsvn/?action=browse&path=%2F*docroot*%2Fhelp_rootview.html
It says

Directory Tarball View - Retrieves a gzipped tar archive containing the contents of a directory.This view is disabled in the default ViewVC configuration.

so this option should be hidden somewhere in the configuration ;)

As I said - it would really be great if you could enable this option - I know that there are definitely some workarounds for this task, but the tar ball option would be the nicest and simplest of all of them.

By the way, thanks for your help!

Greetings
Matthias

Vadim Pisarevsky wrote:

Hi Matthias,

1. The first problem is known, the site admins are working on it.

2. On the tarball option. If you mean stable releases (though, likely you do
not mean it),
they are still available from SourceForge. If you need e.g. nightly tarballs
from the SVN,
I'm not sure if it can be enabled at all, I did not find any place in admin
settings to enable it.

One possible workaround is to use "wget" or other web downloader that allows
recursive retrieval.
Then you may grab the latest version directly from
https://code.ros.org/svn/opencv/trunk/opencv.

For example, if you can open
https://code.ros.org/svn/opencv/trunk/opencv/include/opencv/cxcore.h
in your browser, I do not see any obstacles for getting the whole opencv
tree.

Regards,
Vadim

Re: Problems with the new svn service...

2009-12-07T12:40:47Z

Hi Matthias,

1. The first problem is known, the site admins are working on it.

2. On the tarball option. If you mean stable releases (though, likely you do not mean it),

they are still available from SourceForge. If you need e.g. nightly tarballs from the SVN,

I'm not sure if it can be enabled at all, I did not find any place in admin settings to enable it.

One possible workaround is to use "wget" or other web downloader that allows recursive retrieval.

Then you may grab the latest version directly from

https://code.ros.org/svn/opencv/trunk/opencv.

For example, if you can open

https://code.ros.org/svn/opencv/trunk/opencv/include/opencv/cxcore.h

in your browser, I do not see any obstacles for getting the whole opencv tree.

Regards,

Vadim

On Sat, Dec 5, 2009 at 1:06 PM, hyblade <schmieder.matthias@...> wrote:

Hi Vadim,

1. logging in at https://code.ros.org/gf/ works just fine. But when I browse
to "Trac" to commit or look for bugs, my username and password won't work. I
really don't know what's going wrong...

2. I don't think that the downloading via a svn client will work. Our
corporate fireware is kind of thorough when it comes to file transfers.
Everything is filtered and checked - if there is anything that looks like an
executable it won't pass the check and the hole transfer is denied. So we
have to contact our IT to download such things for us, but as it is the case
they won't use a svn client. The best way for us the get to the actual svn
snapshot was therefore send a link to the tar ball to our IT which
downloaded the files for us.

Is there a reason why you have deactivated the tar ball option? As I said it
would really be great if you re-activate it since there surely are others
that may have the same problem.

Have a nice weekend.

Greetings from Germany!
Matthias

--
View this message in context: http://old.nabble.com/Problems-with-the-new-svn-service...-tp26635800p26654325.html

Sent from the opencvlibrary-devel mailing list archive at Nabble.com.

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Re: Is CvBoostTree::find_split_cat_reg() buggy?

2009-12-07T07:03:14Z

Hi JIA,

is it possible for us to get the training data that you pass to CvBoost (or the .vec file that you pass to traincascade)?

We are not able to reproduce the problem on our current data.

Regards,

Vadim

On Mon, Dec 7, 2009 at 2:06 AM, JIA Pei <jp4work@...> wrote:

Hi, all:

file mlboost.cpp around line 570: sum[i] /= counts[i];

when I debug, I met the situation that counts[i] equals to zero . Is this a bug of current OpenCV trunk?

I notice that it's still ok for me to finish running this program,
however, to be strict, it's better to ensure "divide by 0" will never happen in a strict coding.

Cheers
JIA

--
Welcome to Vision Open
http://www.visionopen.com

Re: Issues with eigenvector computation using OpenCV

2009-12-07T05:46:44Z

Hello,

such reports should better be posted to opencvlibrary-users or submitted as bug reports.

Looking at the code I see that the matrix as declared as "double" and at the same time you build a matrix header with the type CV_32FC1, which means "float".

Regards,

Vadim

On Mon, Dec 7, 2009 at 5:08 AM, Arun A K <arnkrishn@...> wrote:

Hi
I have this matrix A, representing similarities of pixel intensities of an image. For example: Consider a 10 x 10 image. Matrix A in this case would be of dimension 100 x 100, and element A(i,j) would have a value in the range 0 to 1, representing the similarity of pixel i to j in terms of intensity. I am using OpenCV for image processing and the development environment is C on Linux. Objective is to compute the Eigenvectors of matrix A and I have used the following approach: static CvMat mat, *eigenVec, *eigenVal; static double A[100][100]={}, Ain1D[10000]={}; int cnt=0; //Converting matrix A into a one dimensional array //Reason: That is how cvMat requires it for(i = 0;i < affnDim;i++){ for(j = 0;j < affnDim;j++){ Ain1D[cnt++] = A[i][j]; } } mat = cvMat(100, 100, CV_32FC1, Ain1D); cvEigenVV(&mat, eigenVec, eigenVal, 1e-300); for(i=0;i < 100;i++){ val1 = cvmGet(eigenVal,i,0); //Fetching Eigen Value for(j=0;j < 100;j++){ matX[i][j] = cvmGet(eigenVec,i,j); //Fetching each component of Eigenvector i } } Problem: After execution I get nearly all components of all the Eigenvectors to be zero. I tried different images and also tried populating A with random values between 0 and 1, but the same result. Few of the top eigenvalues returned look like the following: 9805401476911479666115491135488.000000 -9805401476911479666115491135488.000000 -89222871725331592641813413888.000000 89222862280598626902522986496.000000 5255391142666987110400.000000 I am now thinking on the lines of using cvSVD() which performs singular value decomposition of real floating-point matrix and might yield me the eigenvectors. But before that I thought of asking it here. Is there anything absurd in my current approach? Am I using the right API i.e. cvEigenVV() for the right input matrix (my matrix A is a floating point matrix)?

Thanks for your time. cheers Arun A K
Graduate Student
Department of Computer Science
Indiana University, Bloomington

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Issues with eigenvector computation using OpenCV

2009-12-06T18:08:53Z

I have this matrix A, representing similarities of pixel intensities of an image. For example: Consider a 10 x 10 image. Matrix A in this case would be of dimension 100 x 100, and element A(i,j) would have a value in the range 0 to 1, representing the similarity of pixel i to j in terms of intensity. I am using OpenCV for image processing and the development environment is C on Linux. Objective is to compute the Eigenvectors of matrix A and I have used the following approach: static CvMat mat, *eigenVec, *eigenVal; static double A[100][100]={}, Ain1D[10000]={}; int cnt=0; //Converting matrix A into a one dimensional array //Reason: That is how cvMat requires it for(i = 0;i < affnDim;i++){ for(j = 0;j < affnDim;j++){ Ain1D[cnt++] = A[i][j]; } } mat = cvMat(100, 100, CV_32FC1, Ain1D); cvEigenVV(&mat, eigenVec, eigenVal, 1e-300); for(i=0;i < 100;i++){ val1 = cvmGet(eigenVal,i,0); //Fetching Eigen Value for(j=0;j < 100;j++){ matX[i][j] = cvmGet(eigenVec,i,j); //Fetching each component of Eigenvector i } } Problem: After execution I get nearly all components of all the Eigenvectors to be zero. I tried different images and also tried populating A with random values between 0 and 1, but the same result. Few of the top eigenvalues returned look like the following: 9805401476911479666115491135488.000000 -9805401476911479666115491135488.000000 -89222871725331592641813413888.000000 89222862280598626902522986496.000000 5255391142666987110400.000000 I am now thinking on the lines of using cvSVD() which performs singular value decomposition of real floating-point matrix and might yield me the eigenvectors. But before that I thought of asking it here. Is there anything absurd in my current approach? Am I using the right API i.e. cvEigenVV() for the right input matrix (my matrix A is a floating point matrix)?

Thanks for your time. cheers Arun A K
Graduate Student
Department of Computer Science
Indiana University, Bloomington

Another mlboost.cpp question...

2009-12-06T16:46:55Z

Hi, All:

In OpenCV 2.x,
file: mlboost.cpp
function: CvBoostTree::find_split_cat_reg
around line: 581

Why using
for( subset_i = 0; subset_i < mi-1; subset_i++ )
but not
for( subset_i = 0; subset_i < mi; subset_i++ )

It looks variable "mi" is the number of categories, but why here "mi-1" is used?
I changed it to "mi" and my testing program runs ok, but can anybody explain why "mi-1" rather than "mi" is used here?

Cheers
JIA

--
Welcome to Vision Open
http://www.visionopen.com

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Welcome to Vision Open
http://www.visionopen.com

Is CvBoostTree::find_split_cat_reg() buggy?

2009-12-06T15:06:22Z

Hi, all:

file mlboost.cpp around line 570: sum[i] /= counts[i];

when I debug, I met the situation that counts[i] equals to zero . Is this a bug of current OpenCV trunk?

I notice that it's still ok for me to finish running this program,
however, to be strict, it's better to ensure "divide by 0" will never happen in a strict coding.

Cheers
JIA

--
Welcome to Vision Open
http://www.visionopen.com

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Welcome to Vision Open
http://www.visionopen.com

CvCascadeBoostTrainData and CvDTreeTrainData documentation...

2009-12-06T11:16:59Z

Hi, all:

A detailed documentation of all classes defined in file "boost.h" is extremely expected.

Let me simply ask some further questions:

In file "boost.cpp" function
void CvCascadeBoostTrainData::setData( const CvFeatureEvaluator* _featureEvaluator,
                                      int _numSamples,
                                      int _precalcValBufSize, int _precalcIdxBufSize,
                                    const CvDTreeParams& _params )
{
..
// around line 271

    var_type->data.i[var_count] = cat_var_count;
    var_type->data.i[var_count+1] = cat_var_count+1;
    work_var_count = ( cat_var_count ? var_count : numPrecalcIdx ) + 1;
    buf_size = (work_var_count + 1) * sample_count;
    buf_count = 2;

}

In fact, all the member variables within both class CvCascadeBoostTrainData and CvDTreeTrainData are used in this function.
All member variables are pretty important but unfortunately, there is no documentation on these variables.
Particularly, for class CvDTreeTrainData, there are so many member variables as:

    vector<vector<float> > pred_float_buf;
    vector<vector<int> > pred_int_buf;
    vector<vector<float> > resp_float_buf;
    vector<vector<int> > resp_int_buf;
    vector<vector<int> > cv_lables_buf;
    vector<vector<int> > sample_idx_buf;

    int sample_count, var_all, var_count, max_c_count;
    int ord_var_count, cat_var_count, work_var_count;
    bool have_labels, have_priors;
    bool is_classifier;
    int tflag;

    const CvMat* train_data;
    const CvMat* responses;
    CvMat* responses_copy; // used in Boosting

    int buf_count, buf_size;
    bool shared;
    int is_buf_16u;

    CvMat* cat_count;
    CvMat* cat_ofs;
    CvMat* cat_map;

    CvMat* counts;
    CvMat* buf;
    CvMat* direction;
    CvMat* split_buf;

    CvMat* var_idx;
    CvMat* var_type; // i-th element =
                     //   k<0 - ordered
                     //   k>=0 - categorical, see k-th element of cat_* arrays
    CvMat* priors;
    CvMat* priors_mult;

    CvDTreeParams params;

    CvMemStorage* tree_storage;
    CvMemStorage* temp_storage;

    CvDTreeNode* data_root;

    CvSet* node_heap;
    CvSet* split_heap;
    CvSet* cv_heap;
    CvSet* nv_heap;

    CvRNG rng;

Can anybody please explain the purposes of all these variables?

Best Regards
JIA Pei

--
Welcome to Vision Open
http://www.visionopen.com

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Welcome to Vision Open
http://www.visionopen.com

What is the purpose of get_work_var_count() for class CvDTreeTrainData?

2009-12-05T16:23:17Z

Hi, Dear all:

What is the purpose of the function
int get_work_var_count() const {return work_var_count;}
in class
struct CV_EXPORTS CvDTreeTrainData
???

We may notice that in file "boost.cpp", there are the following 2 functions defined around line 376-385

void CvCascadeBoostTrainData::get_sample_indices( CvDTreeNode* n, int* indicesBuf, const int** indices )
{
CvDTreeTrainData::get_cat_var_data( n, get_work_var_count(), indicesBuf, indices );
}

void CvCascadeBoostTrainData::get_cv_labels( CvDTreeNode* n, int* labels_buf, const int** labels )
{
CvDTreeTrainData::get_cat_var_data( n, get_work_var_count()- 1, labels_buf, labels );
}

But, why the first function uses get_work_var_count() while the second function uses get_work_var_count()-1 ????

It looks for me that get_work_var_count()-1 is exactly the number of extracted features, right?
But, what does get_work_var_count() represent for?

Seriously can't understand this part of code and a detailed explanation is extremely and urgently expected.

Best Regards
JIA Pei

--
Welcome to Vision Open
http://www.visionopen.com

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Welcome to Vision Open
http://www.visionopen.com

Again -- traincascade question

2009-12-05T14:06:06Z

Hi, all:

I'm still testing traincascade of OpenCV 2.x .

update_weights( 0 );
of function
bool CvCascadeBoost::train( const CvFeatureEvaluator* _featureEvaluator,
                           int _numSamples,
                           int _precalcValBufSize, int _precalcIdxBufSize,
                           const CvCascadeBoostParams& _params )
in file "boost.cpp" leads me to function "update_weights" in the same file.

void CvCascadeBoost::update_weights( CvBoostTree* tree )
{
    int n = data->sample_count;
    double sumW = 0.;
    int step = 0;
    float* fdata = 0;
    int *sampleIdxBuf;
    const int* sampleIdx = 0;
    if ( (params.boost_type == LOGIT) || (params.boost_type == GENTLE) )
    {
        step = CV_IS_MAT_CONT(data->responses_copy->type) ?
            1 : data->responses_copy->step / CV_ELEM_SIZE(data->responses_copy->type);
        fdata = data->responses_copy->data.fl;
        sampleIdxBuf = (int*)cvStackAlloc(data->sample_count*sizeof(sampleIdxBuf[0]));
        data->get_sample_indices( data->data_root, sampleIdxBuf, &sampleIdx );
    }
....
}

I tried to print out the following several variables' values, but I couldn't understand what's happening inside ( I revised the above code a bit to the following)

    if ( (params.boost_type == LOGIT) || (params.boost_type == GENTLE) )
    {
        cout << data->response_copy->type << endl;
        cout << data->response_copy->step << endl;
        cout << CV_ELEM_SIZE(data->responses_copy->type) << endl;
        cout << CV_IS_MAT_CONT(data->responses_copy->type) << endl;

        step = CV_IS_MAT_CONT(data->responses_copy->type) ?
            1 : data->responses_copy->step / CV_ELEM_SIZE(data->responses_copy->type);
        fdata = data->responses_copy->data.fl;
        sampleIdxBuf = (int*)cvStackAlloc(data->sample_count*sizeof(sampleIdxBuf[0]));
        data->get_sample_indices( data->data_root, sampleIdxBuf, &sampleIdx );
    }

and I obtained the values as:

1111638021
4
4
16384

This is telling that data->response_copy->type is 1111638021 . What type is it for this matrix?
What is the function CV_IS_MAT_CONT(data->responses_copy->type) for??
Why for this function, 16384 is printed out?

Can anybody please give me a detailed explanation?

Best Regards
JIA

--
Welcome to Vision Open
http://www.visionopen.com

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Welcome to Vision Open
http://www.visionopen.com

Re: Problems with the new svn service...

2009-12-05T02:06:53Z

Hi Vadim,

1. logging in at https://code.ros.org/gf/ works just fine. But when I browse to "Trac" to commit or look for bugs, my username and password won't work. I really don't know what's going wrong...

2. I don't think that the downloading via a svn client will work. Our corporate fireware is kind of thorough when it comes to file transfers. Everything is filtered and checked - if there is anything that looks like an executable it won't pass the check and the hole transfer is denied. So we have to contact our IT to download such things for us, but as it is the case they won't use a svn client. The best way for us the get to the actual svn snapshot was therefore send a link to the tar ball to our IT which downloaded the files for us.

Is there a reason why you have deactivated the tar ball option? As I said it would really be great if you re-activate it since there surely are others that may have the same problem.

Have a nice weekend.

Greetings from Germany!
Matthias

Re: Python interface and Ipython

2009-12-04T12:48:16Z

Jorge,

Thank you! I did not realize that you use the new Python interface.

I will apply your patch. Thank you for exploring and solving this!

Regards,

Vadim

On Fri, Dec 4, 2009 at 11:43 PM, Jorge Scandaliaris <jorgesmbox-ml@...> wrote:

Vadim Pisarevsky <vadim.pisarevsky@...> writes:

<...>

> In Linux, where highgui is built on GTK+ 2.x, there is a way to create a
separate thread that will manage window events:
> cvStartWindowThread();
> I think, it should be available from Python. Try to call it.
>

StartWindowThread() is in the source tree but not in the new python interface.
Fortunately, after a bit of messing around with it I was able to add it, with 1
line of code! I didn't know exactly what I was doing, but it seems to work now.
Maybe there is a good reason to leave this function out of the python interface,
but as I said, it seems to work here.
The change was to add a line with the name of the function in
interfaces/python/api and re-build.
Thanks very much vadim for the pointer!

Regards,

Jorge

Index: opencv/interfaces/python/api
===================================================================
--- opencv/interfaces/python/api (revisión: 2281)
+++ opencv/interfaces/python/api (copia de trabajo)
@@ -1532,7 +1532,7 @@
DecodeImageM CvMat*
CvMat buf
int iscolor CV_LOAD_IMAGE_COLOR
-
+StartWindowThread
# cvaux stuff
HOGDetectMultiScale CvSeq*
CvArr image

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Re: Python interface and Ipython

2009-12-04T12:43:07Z

Vadim Pisarevsky <vadim.pisarevsky@...> writes:

<...>
> In Linux, where highgui is built on GTK+ 2.x, there is a way to create a
separate thread that will manage window events:
> cvStartWindowThread();
> I think, it should be available from Python. Try to call it.
>

StartWindowThread() is in the source tree but not in the new python interface.
Fortunately, after a bit of messing around with it I was able to add it, with 1
line of code! I didn't know exactly what I was doing, but it seems to work now.
Maybe there is a good reason to leave this function out of the python interface,
but as I said, it seems to work here.
The change was to add a line with the name of the function in
interfaces/python/api and re-build.
Thanks very much vadim for the pointer!

Regards,

Jorge

Index: opencv/interfaces/python/api
===================================================================
--- opencv/interfaces/python/api (revisión: 2281)
+++ opencv/interfaces/python/api (copia de trabajo)
@@ -1532,7 +1532,7 @@
DecodeImageM CvMat*
CvMat buf
int iscolor CV_LOAD_IMAGE_COLOR
-
+StartWindowThread
# cvaux stuff
HOGDetectMultiScale CvSeq*
CvArr image

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

How does CvCascadeClassifier::train work?

2009-12-04T12:10:01Z

Hi, all:

In file ...\opencv\apps\traincascade\cascadeclassifier.cpp
how does CvCascadeClassifier::train work?
Particularly, what is the purpose of the code between
//////////////////////////////////////////////////////////////////////////////////////////// ?

In my testing, I selected "numStages == 14", but
the for loop "for( int i = startNumStages; i < numStages; i++ )" jumped out
when "i==6", at line
" if ( !fs.isOpened() )                                                                          // where the loops jumped out.
            return false; "

It looks like the entire training process doesn't need 14 stages that I specified, but converged just within 6 stages,
and produced 6 .xml files "stage0.xml", "stage1.xml", "stage2.xml", "stage3.xml", "stage4.xml", "stage5.xml".

A detailed manual describing how traincascade is working is extremely urgently expected.

bool CvCascadeClassifier::train( const String _cascadeDirName,
                                const String _posFilename,
                                const String _negFilename,
                                int _numPos, int _numNeg,
                                int _precalcValBufSize, int _precalcIdxBufSize,
                                int _numStages,
                                const CvCascadeParams& _cascadeParams,
                                const CvFeatureParams& _featureParams,
                                const CvCascadeBoostParams& _stageParams,
                                bool baseFormatSave )
{
...
////////////////////////////////////////////////////////////////////////////////////////////
    for( int i = startNumStages; i < numStages; i++ )
    {
        cout << endl << "===== TRAINING " << i << "-stage =====" << endl;
        cout << "<BEGIN" << endl;

        if ( !updateTrainingSet( tempLeafFARate ) )
        {
            cout << "Train dataset for temp stage can not be filled."
                "Branch training terminated." << endl;
            break;
        }
        if( tempLeafFARate <= requiredLeafFARate )
        {
            cout << "Required leaf false alarm rate achieved. "
                 "Branch training terminated." << endl;
            break;
        }

        CvCascadeBoost* tempStage = new CvCascadeBoost;
        tempStage->train( (CvFeatureEvaluator*)featureEvaluator,
                           curNumSamples, _precalcValBufSize, _precalcIdxBufSize,
                          *((CvCascadeBoostParams*)stageParams) );
        stageClassifiers.push_back( tempStage );

        cout << "END>" << endl;

        // save params
        String filename;
        if ( i == 0)
        {
            filename = dirName + CC_PARAMS_FILENAME;
            FileStorage fs( filename, FileStorage::WRITE);
            if ( !fs.isOpened() )
                return false;
            fs << FileStorage::getDefaultObjectName(filename) << "{";
            writeParams( fs );
            fs << "}";
        }
        // save temp stage
        char buf[10];
        sprintf(buf, "%s%d", "stage", i );
        filename = dirName + buf + ".xml";
        FileStorage fs( filename, FileStorage::WRITE );
        if ( !fs.isOpened() )                                                                          // where the loops jumped out.
            return false;
        fs << FileStorage::getDefaultObjectName(filename) << "{";
        tempStage->write( fs, Mat() );
        fs << "}";
    }
////////////////////////////////////////////////////////////////////////////////////////////
...
}

Best Regards
JIA

--
Welcome to Vision Open
http://www.visionopen.com

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Welcome to Vision Open
http://www.visionopen.com

Re: Python interface and Ipython

2009-12-04T10:31:08Z

Vadim Pisarevsky <vadim.pisarevsky@...> writes:

> what's the OS?
> In Linux, where highgui is built on GTK+ 2.x, there is a way to create a
separate thread that will manage window events:
> cvStartWindowThread();
> I think, it should be available from Python. Try to call it.

Yes, it's Linux. I can't seem to find the function you mention, though. Not in
the python interface nor in any of the docs. Maybe it got dropped when moving
from 1.1 to 2.0? This is OpenCV-2.0, not svn.
>
> Actually, highgui window system has never been properly tested in interactive
environments. That is, if it works well in a standard Python shell, that's
already great news :)

It didn't seem good news to me at the time, but considering what you say, I
should probably be happy now.

> We normally just run small programs in C/C++ or Python and see what's
happening in debugger, or add printf statements.

Well, me too, all the time, but images are a bit of pain to debug that with
printfs ;) Jokes aside, I am really glad I have libraries like opencv which make
my work a lot easier, even if sometimes I find some bumps on the road.

Regards,

Jorge

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Re: Python interface and Ipython

2009-12-04T09:47:07Z

Hi,

what's the OS?

In Linux, where highgui is built on GTK+ 2.x, there is a way to create a separate thread that will manage window events:

cvStartWindowThread();

I think, it should be available from Python. Try to call it.

Actually, highgui window system has never been properly tested in interactive environments. That is, if it works well in a standard Python shell, that's already great news :)

We normally just run small programs in C/C++ or Python and see what's happening in debugger, or add printf statements.

Regards,

Vadim

On Fri, Dec 4, 2009 at 7:57 PM, Jorge Scandaliaris <jorgesmbox-ml@...> wrote:

Hi,
I just started using opencv-2.0 from the python interface. I use Ipython for all
my python work, so that's what I used with opencv. There is strange issue,
however, when using Ipython and highgui features like windows. They appear ok
but they won't close later on. I manage to open and close windows as long as I
don't call cv.WaitKey(). In the standard python shell, all seems fine, although
I just made a couple of tests.
Does anyone using the python interface experienced this? Any ideas of what
causes this and how to fix it/avoid it? If I have to, I'll switch to python
shell, but it makes everything harder.

Regards,

Jorge

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Python interface and Ipython

2009-12-04T08:57:41Z

Hi,
I just started using opencv-2.0 from the python interface. I use Ipython for all
my python work, so that's what I used with opencv. There is strange issue,
however, when using Ipython and highgui features like windows. They appear ok
but they won't close later on. I manage to open and close windows as long as I
don't call cv.WaitKey(). In the standard python shell, all seems fine, although
I just made a couple of tests.
Does anyone using the python interface experienced this? Any ideas of what
causes this and how to fix it/avoid it? If I have to, I'll switch to python
shell, but it makes everything harder.

Regards,

Jorge

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Re: Problems with the new svn service...

2009-12-04T06:10:23Z

Hi!

1. Did you try to login directly at https://code.ros.org/gf/ ?

May be this way it will work.

2. For the code you do not have to use ViewVC,

you may point your svn client directly to https://code.ros.org/svn/opencv/trunk.

I'm not behind a corporate firewall (though there is local firewall on my machine) and checkouts/checkins proceed fine.

So it should be fine on your machine too, including the anonymous access. In fact, I'm only prompted for a password when I commit things.

On Monday I can ask my friend to try to load the latest opencv from within a corporate network.

Regards,

Vadim

On Fri, Dec 4, 2009 at 10:22 AM, hyblade <schmieder.matthias@...> wrote:

I just tried to get a acquainted with the new service but failed enormously.
I
tried to summit a bug by following the link posted in the wiki
(https://code.ros.org/trac/opencv). After that I followed the text on the
homepage and registered at http://code.ros.org, which by the way seems like
a
really nice page for all of our needs! After completing the registration
process
by clicking the link that comes with the registration email I tried to log
me in
at https://code.ros.org/trac/opencv. It failed every time I tried – my
username
and password only works on http://code.ros.org. This should probably be the
problem way I can't commit a bug. The text on the homepage says:

"Click on "new ticket" to file a new bug report or feature request, or
submit a
patch."

But this "new ticket" isn't shown for me ;(

Could you please tell me how to log in there so I can commit my thoughts?!

There's also something else. With the old svn service there had always been
the
possibility to download the latest svn snapshot as a tar ball. Since you're
using ViewVC this should still be possible. I would be great if you would
(re)enable this option because some of us out there can't get beyond our
firewalls with svn clients.

THX!
--
View this message in context: http://old.nabble.com/Problems-with-the-new-svn-service...-tp26635800p26635800.html
Sent from the opencvlibrary-devel mailing list archive at Nabble.com.

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

How does "traincascade" work?

2009-12-04T04:30:06Z

Hi, sorry for my disturbance guys.

I'm trying to train my own cascade files for face detection by using OpenCV application "traincascade".
It's seriously painful that I can't figure out how "traincascade" works.

In the function

bool CvCascadeClassifier::train( const String _cascadeDirName,
                                const String _posFilename,
                                const String _negFilename,
                                int _numPos, int _numNeg,
                                int _precalcValBufSize, int _precalcIdxBufSize,
                                int _numStages,
                                const CvCascadeParams& _cascadeParams,
                                const CvFeatureParams& _featureParams,
                                const CvCascadeBoostParams& _stageParams,
                                bool baseFormatSave )

if ( !load( dirName ) ) // around line 148 of file..\opencv\apps\traincascade\cascadeclassifier.cpp

Seriously can't understand thoroughly what has been loaded by the function load() ???

In file ..\opencv\apps\traincascade\cascadeclassifier.cpp

FileNode node = fs.getFirstTopLevelNode();
seems to be very important for the following function
bool CvCascadeClassifier::load( const String cascadeDirName )
{
    FileStorage fs( cascadeDirName + CC_PARAMS_FILENAME, FileStorage::READ );
    if ( !fs.isOpened() )
        return false;
    FileNode node = fs.getFirstTopLevelNode();

...
}

We can see that

bool CvCascadeBoost::read( const FileNode &node,
                           const CvFeatureEvaluator* _featureEvaluator,
                           const CvCascadeBoostParams& _params )
{
...
node[CC_STAGE_THRESHOLD] >> threshold;
...
}
threshold was assigned by node[CC_STAGE_THRESHOLD] .

Therefore, "node" must be first assigned in function
bool CvCascadeClassifier::load( const String cascadeDirName )
by FileNode node = fs.getFirstTopLevelNode();

I'm just wondering where and when "node" has been assigned the value "node[CC_STAGE_THRESHOLD]", for later threshold assignment?

Since node[CC_STAGE_THRESHOLD] has been assigned some value, I'm wondering does this assignment contain some training inside?
Because, obviously, we can't randomly assign a threshold. Threshold must be a well-designed/well-trained one.

So, can anybody please help on how, when and where node[CC_STAGE_THRESHOLD] is assigned?

Best Regards
JIA

--
Welcome to Vision Open
http://www.visionopen.com

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Welcome to Vision Open
http://www.visionopen.com

Problems with the new svn service...

2009-12-03T23:22:29Z

I just tried to get a acquainted with the new service but failed enormously. I
tried to summit a bug by following the link posted in the wiki
(https://code.ros.org/trac/opencv). After that I followed the text on the
homepage and registered at http://code.ros.org, which by the way seems like a
really nice page for all of our needs! After completing the registration process
by clicking the link that comes with the registration email I tried to log me in
at https://code.ros.org/trac/opencv. It failed every time I tried – my username
and password only works on http://code.ros.org. This should probably be the
problem way I can't commit a bug. The text on the homepage says:

"Click on "new ticket" to file a new bug report or feature request, or submit a
patch."

But this "new ticket" isn't shown for me ;(

Could you please tell me how to log in there so I can commit my thoughts?!

There's also something else. With the old svn service there had always been the
possibility to download the latest svn snapshot as a tar ball. Since you're
using ViewVC this should still be possible. I would be great if you would
(re)enable this option because some of us out there can't get beyond our
firewalls with svn clients.

THX!

Re: Win64 build errors in SVN and proposed resolution

2009-12-03T06:58:53Z

> Your patches are applied in SVN 2353
Thank you! I can confirm that r2353 builds without any workarounds on
the above-mentioned configuration.

Best Regards
Jan Wassenberg

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Re: Win64 build errors in SVN and proposed resolution

2009-12-03T06:36:49Z

Thanks!

Your patches are applied in SVN 2353

Regards,

Vadim

On Wed, Dec 2, 2009 at 6:51 PM, Jan Wassenberg <jan.wassenberg@...> wrote:

Dear developers,

I'd like to report x64 build issues and a possible resolution. Steps taken:
. checkout from svn on 2009-12-02T14:37:00
. configure with default options;
. generate VC2008 project for Win64, native compiler;
. build entire solution in debug mode.

Building src\cxcore\cxsystem.cpp fails (MSVC doesn't allow inline asm
in x64 builds, but there's _asm _emit for RDTSC because WIN32 is
defined).
Proposed resolution is:
-#elif defined _MSC_VER && defined WIN32
+#elif defined _MSC_VER && defined WIN32 && !defined _WIN64
(MSDN says _WIN32 and _WIN64 are predefined macros, and _WIN32 is also
automatically defined in 64-bit builds, which causes your WIN32 macro
to be defined by _cxcore.h) It may be preferable to change the
definition of WIN32 if you are intending that it really mean
32-bit-only Windows (unlike the predefined macro).

Another issue is due to IPP, I am using the x64 version packaged with
ICC 11.0.075 Professional. The linker is complaining that
"ippcorelem64t.lib" isn't found. The second directory in the
CMake-generated project linker path correctly points to the IPP lib
directory, but that specified file doesn't exist. Instead, the only
ippcore*.lib are ippcoreem64tl.lib and ippcoreem64t_t.lib . Looking at
the other referenced libraries, it does not appear to be the intent to
use *t_t (static linking without dispatching), so I've replaced all
instances of ippcorelem64t.lib in every .vcproj with ippcoreem64tl.lib
.

After this, all projects except ffopencv (which is apparently missing
FFPMEG, I don't have it installed) build.

Hope this helps.
Cheers
Jan Wassenberg

———————————————————————————————————————————————————————————
Dipl.-Inform. Jan Wassenberg
Fraunhofer FOM
Forschungsinstitut für Optronik und Mustererkennung Gutleuthausstr. 1,
76275 Ettlingen, Germany Telefon +49 7243 992-319 Fax +49 7243 992-299
wassenberg@... www.fom.fgan.de
———————————————————————————————————————————————————————————

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Re: About OpenCV 2.x Decision Tree

2009-12-03T01:42:01Z

Hello,

We are working on improving documentation of the newly added functionality.

CvCascadeBoostTree does not support trimming, and it does not need it probably. In boosting (as well as in random trees and extremely random trees) the idea is to have many-many decision trees, where each tree is quite small, usually 1 or 2 levels (not counting leaves). Such trees do not need to be pruned, so this functionality is not implemented in OpenCV.

Regards,

Vadim

On Wed, Dec 2, 2009 at 5:35 AM, JIA Pei <jp4work@...> wrote:

Hi, all:

It looks like the current OpenCV 2.x downloaded from https://code.ros.org/svn/opencv/trunk/
doesn't actually afford an up-to-date documentation.

For example, I couldn't find any documentation about
CvCascadeBoostTree etc.

Is there any documentation about how OpenCV's decision tree works? How does it determine to carry out trimming?

Best Regards
JIA

--
Welcome to Vision Open
http://www.visionopen.com

Win64 build errors in SVN and proposed resolution

2009-12-02T07:51:02Z

Dear developers,

I'd like to report x64 build issues and a possible resolution. Steps taken:
. checkout from svn on 2009-12-02T14:37:00
. configure with default options;
. generate VC2008 project for Win64, native compiler;
. build entire solution in debug mode.

Building src\cxcore\cxsystem.cpp fails (MSVC doesn't allow inline asm
in x64 builds, but there's _asm _emit for RDTSC because WIN32 is
defined).
Proposed resolution is:
-#elif defined _MSC_VER && defined WIN32
+#elif defined _MSC_VER && defined WIN32 && !defined _WIN64
(MSDN says _WIN32 and _WIN64 are predefined macros, and _WIN32 is also
automatically defined in 64-bit builds, which causes your WIN32 macro
to be defined by _cxcore.h) It may be preferable to change the
definition of WIN32 if you are intending that it really mean
32-bit-only Windows (unlike the predefined macro).

Another issue is due to IPP, I am using the x64 version packaged with
ICC 11.0.075 Professional. The linker is complaining that
"ippcorelem64t.lib" isn't found. The second directory in the
CMake-generated project linker path correctly points to the IPP lib
directory, but that specified file doesn't exist. Instead, the only
ippcore*.lib are ippcoreem64tl.lib and ippcoreem64t_t.lib . Looking at
the other referenced libraries, it does not appear to be the intent to
use *t_t (static linking without dispatching), so I've replaced all
instances of ippcorelem64t.lib in every .vcproj with ippcoreem64tl.lib
.

After this, all projects except ffopencv (which is apparently missing
FFPMEG, I don't have it installed) build.

Hope this helps.
Cheers
Jan Wassenberg

———————————————————————————————————————————————————————————
Dipl.-Inform. Jan Wassenberg
Fraunhofer FOM
Forschungsinstitut für Optronik und Mustererkennung Gutleuthausstr. 1,
76275 Ettlingen, Germany Telefon +49 7243 992-319 Fax +49 7243 992-299
wassenberg@... www.fom.fgan.de
———————————————————————————————————————————————————————————

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

OpenCV R2348 DOC won't compile...

2009-12-01T20:30:18Z

Hi, all:

I noticed that OpenCV 2.x is dramatically updating the documentation, which is absolutely what I expected.
However, I'm not able to compile DOC of OpenCV R2348. Just wonder when the most up-to-date documentation will be ready??

[ 79%] Generating CxCore.tex
cd /home/jiapei/Tools/CV/opencv-svn-R2348/opencv/doc && /usr/bin/cmake -E copy CxCore.tex /home/jiapei/Tools/CV/opencv-svn-R2348/opencv/build/doc/CxCore.tex
Error copying file "CxCore.tex" to "/home/jiapei/Tools/CV/opencv-svn-R2348/opencv/build/doc/CxCore.tex".
make[2]: *** [doc/CxCore.tex] Error 1
make[2]: Leaving directory `/home/jiapei/Tools/CV/opencv-svn-R2348/opencv/build'
make[1]: *** [doc/CMakeFiles/ReferenceManual.dir/all] Error 2
make[1]: Leaving directory `/home/jiapei/Tools/CV/opencv-svn-R2348/opencv/build'
make: *** [all] Error 2

Best Regards
JIA

--
Welcome to Vision Open
http://www.visionopen.com

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Welcome to Vision Open
http://www.visionopen.com

About OpenCV 2.x Decision Tree

2009-12-01T18:35:47Z

Hi, all:

It looks like the current OpenCV 2.x downloaded from https://code.ros.org/svn/opencv/trunk/
doesn't actually afford an up-to-date documentation.

For example, I couldn't find any documentation about
CvCascadeBoostTree etc.

Is there any documentation about how OpenCV's decision tree works? How does it determine to carry out trimming?

Best Regards
JIA

--
Welcome to Vision Open
http://www.visionopen.com

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Welcome to Vision Open
http://www.visionopen.com

Re: blas opencl

2009-12-01T13:16:59Z

Hello Stafano,

While scanning through my e-mail I noticed this post that I somehow forgot to reply to. Sorry.

We are looking at using GPUs for OpenCV acceleration, and OpenCL though probably be the best API.

On the other hand, optimizing BLAS with OpenCL is not that interesting for us - we'd better have faster stereo correspondence or faster object tracking or another computer-vision-specific algorithm. That is, that's great of course if GPU can compute 1000x1000 matrix product much faster than CPU, but it does not matter much for us until we figure out how to solve our practical problems using this feature.

Regards,

Vadim

On Thu, Nov 26, 2009 at 1:58 AM, bluefuture@... <bluefuture@...> wrote:

Hello Vadim,
is there any plan or interest about opencl support in opencv?
I see that there is an initial blas opencl code on sourceforge[¹].

[¹]http://sourceforge.net/projects/openclblas/

Cheers,
Stefano

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
trial. Simplify your report design, integration and deployment - and focus on
what you do best, core application coding. Discover what's new with
Crystal Reports now. http://p.sf.net/sfu/bobj-july
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

my email id

2009-12-01T12:43:40Z

ankit.096@...
------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing.
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Re: patch for JPEG errors

2009-12-01T12:02:26Z

Hi developers,

My name is Ankit and I am new to openCV development and hope to learn a lot in this forum.
I need to the patch file (or any other way :) ) to fix the jpeg decode errors in opencv. The error is -
"Corrupt JPEG data: premature end of data segment" .
I am using latest version of ubuntu 9.10 .
Also can you please let me know how to install it. My email id is ankit.096@gmail.com

Thanks,
Ankit

Mikhail-11 wrote:

Hello,

I have discovered the bug in OpenCV from CVS, on Linux. When reading data
from V4L2 cameras, the buffer is released too early and therefore sometimes it
gets overridden. This is probably OK for non-compressed camera, but it causes
JPEG decode errors and horrible image damage on JPEG cameras (like low-cost
2-megapixels USB webcams from Logitech/Microsoft).

The attached patch fixes the following:
(1) When reading from camera, JPEG decode warnings become decode errors
(2) JPEG decode errors are handled by the camera, thus user-app can detect
damaged data
(3) Better buffer management ensures that the data will not be damaged in
the first place.

Mikhail

Index: ./otherlibs/highgui/cvcap_v4l.cpp
===================================================================
RCS file: /cvsroot/opencvlibrary/opencv/otherlibs/highgui/cvcap_v4l.cpp,v
retrieving revision 1.64
diff -u -r1.64 cvcap_v4l.cpp
--- ./otherlibs/highgui/cvcap_v4l.cpp 9 May 2008 06:01:52 -0000 1.64
+++ ./otherlibs/highgui/cvcap_v4l.cpp 2 Jul 2008 08:51:48 -0000
@@ -146,6 +146,15 @@
some drivers working, but not fully with V4L2. (so, we do not know when we
need to switch from V4L2 to V4L.

+10th patch: July 02, 2008, Mikhail Afanasyev fopencv@theamk.com
+Fix reliability problems with high-resolution UVC cameras on linux
+the symptoms were damaged image and 'Corrupt JPEG data: premature end of data segment' on stderr
+- V4L_ABORT_BADJPEG detects JPEG warnings and turns them into errors, so bad images
+ could be filtered out
+- USE_TEMP_BUFFER fixes the main problem (improper buffer management) and
+ prevents bad images in the first place
+
+
make & enjoy!

*/
@@ -225,6 +234,18 @@
#define CHANNEL_NUMBER 1
#define MAX_CAMERAS 8

+
+// default and maximum number of V4L buffers, not including last, 'special' buffer
+#define MAX_V4L_BUFFERS 10
+#define DEFAULT_V4L_BUFFERS 4
+
+// if enabled, copies data from the buffer. this uses a bit more memory,
+// but much more reliable for some UVC cameras
+#define USE_TEMP_BUFFER
+
+// if enabled, then bad JPEG warnings become errors and cause NULL returned instead of image
+#define V4L_ABORT_BADJPEG
+
#define MAX_DEVICE_DRIVER_NAME 80

/* Device Capture Objects */
@@ -275,7 +296,7 @@
#ifdef HAVE_CAMV4L2

/* V4L2 variables */
- buffer buffers[10];
+ buffer buffers[MAX_V4L_BUFFERS + 1];
struct v4l2_capability cap;
struct v4l2_input inp;
struct v4l2_format form;
@@ -862,7 +883,7 @@

CLEAR (capture->req);

- unsigned int buffer_number = 4;
+ unsigned int buffer_number = DEFAULT_V4L_BUFFERS;

try_again:

@@ -894,6 +915,7 @@
return -1;
} else {
buffer_number--;
+ fprintf (stderr, "Insufficient buffer memory on %s -- decreaseing buffers\n", deviceName);

goto try_again;
}
@@ -932,6 +954,13 @@
icvCloseCAM_V4L (capture);
return -1;
}
+
+#ifdef USE_TEMP_BUFFER
+ if (n_buffers == 0) {
+ capture->buffers[MAX_V4L_BUFFERS].start = malloc( buf.length );
+ capture->buffers[MAX_V4L_BUFFERS].length = buf.length;
+ };
+#endif
}

/* Set up Image data */
@@ -1155,7 +1184,16 @@

assert(buf.index < capture->req.count);

+#ifdef USE_TEMP_BUFFER
+ memcpy(capture->buffers[MAX_V4L_BUFFERS].start,
+ capture->buffers[buf.index].start,
+ capture->buffers[MAX_V4L_BUFFERS].length );
+ capture->bufferIndex = MAX_V4L_BUFFERS;
+ //printf("got data in buff %d, len=%d, flags=0x%X, seq=%d, used=%d)\n",
+ // buf.index, buf.length, buf.flags, buf.sequence, buf.bytesused);
+#else
capture->bufferIndex = buf.index;
+#endif

if (-1 == xioctl (capture->deviceHandle, VIDIOC_QBUF, &buf))
perror ("VIDIOC_QBUF");
@@ -1265,6 +1303,12 @@
}

}
+
+#ifdef V4L_ABORT_BADJPEG
+ // skip first frame. it is often bad -- this is unnotied in traditional apps,
+ // but could be fatal if bad jpeg is enabled
+ mainloop_v4l2(capture);
+#endif

/* preparation is ok */
capture->FirstCapture = 0;
@@ -1692,6 +1736,19 @@
longjmp( err_mgr->setjmp_buffer, 1 );
}

+#ifdef V4L_ABORT_BADJPEG
+void emit_message (j_common_ptr cinfo, int msg_level) {
+ char buffer[JMSG_LENGTH_MAX];
+ GrFmtJpegErrorMgr* err_mgr = (GrFmtJpegErrorMgr*)(cinfo->err);
+ if (msg_level >= 0) {
+ return;
+ };
+ (*cinfo->err->format_message) (cinfo, buffer);
+ fprintf(stderr, "Camera decode error (%d): %s\n", msg_level, buffer);
+ longjmp( err_mgr->setjmp_buffer, 1 );
+};
+#endif
+
/////////////////////// MyMJpegReader ///////////////////

/* constructor just call the parent constructor, but without real filename */
@@ -1715,6 +1772,9 @@

cinfo->err = jpeg_std_error(&jerr->pub);
jerr->pub.error_exit = error_exit;
+#ifdef V4L_ABORT_BADJPEG
+ jerr->pub.emit_message = emit_message;
+#endif

m_cinfo = cinfo;
m_jerr = jerr;
@@ -1744,18 +1804,27 @@
}

/* convert from mjpeg to rgb24 */
-static void
+static bool
mjpeg_to_rgb24 (int width, int height,
unsigned char *src, int length,
unsigned char *dst)
{
/* use a MyMJpegReader reader for doing the conversion */
MyMJpegReader* reader = 0;
+ bool ok;
reader = new MyMJpegReader (src, length);
- reader->ReadHeader ();
- reader->ReadData (dst, width * 3, 1 );
+ ok = reader->ReadHeader ();
+ if (ok) {
+ ok = reader->ReadData (dst, width * 3, 1 );
+ };
delete reader;
-
+ return ok;
+ /*
+ if (!ok) {
+ fprintf(stderr, "Camera returned bad header\n");
+ memset(dst, 0, width * 3 * height);
+ };
+ */
}

#endif
@@ -2092,12 +2161,13 @@
because it's use libjepg and fmemopen()
*/
if (PALETTE_MJPEG == 1)
- mjpeg_to_rgb24(capture->form.fmt.pix.width,
- capture->form.fmt.pix.height,
- (unsigned char*)(capture->buffers[capture->bufferIndex]
- .start),
- capture->buffers[capture->bufferIndex].length,
- (unsigned char*)capture->frame.imageData);
+ if (!mjpeg_to_rgb24(capture->form.fmt.pix.width,
+ capture->form.fmt.pix.height,
+ (unsigned char*)(capture->buffers[capture->bufferIndex]
+ .start),
+ capture->buffers[capture->bufferIndex].length,
+ (unsigned char*)capture->frame.imageData))
+ return 0;
#endif
#endif

-------------------------------------------------------------------------
Sponsored by: SourceForge.net Community Choice Awards: VOTE NOW!
Studies have shown that voting for your favorite open source project,
along with a healthy diet, reduces your potential for chronic lameness
and boredom. Vote Now at http://www.sourceforge.net/community/cca08
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Re: OpenCV rev 2336 build failed in MinGW 4.4

2009-11-30T06:32:53Z

Nicolas Chauvet wrote:

2009/11/30 Asm_gmail asmwarrior@...:

I try to build the latest SVN code under TDM-GCC 4.4.1 and Cmake 2.8.0

There is the build error, it seems the type is not declared it seems
this is introduces in the commit 2336.

Building CXX object 3rdparty/libtiff/CMakeFiles/libtiff.dir/tif_stream.obj
cd F:\test1\3rdparty\libtiff && D:\MinGW\bin\g++.exe   -DHAVE_CONFIG_H
-Wall -ffunction-sections  -g  -O0 -DDEBUG -D_DEBUG  -IF:\opencv\.
-IF:\test1 -IF:\opencv\include -IF:\opencv\include\opencv
-IF:\opencv\3rdparty\libtiff -IF:\opencv\3rdparty\libtiff\..\include
-o CMakeFiles\libtiff.dir\tif_stream.obj -c
F:\opencv\3rdparty\libtiff\tif_stream.cxx
In file included from F:\opencv\3rdparty\libtiff\tiffio.h:33,
                from F:\opencv\3rdparty\libtiff\tiffiop.h:53,
                from F:\opencv\3rdparty\libtiff\tif_stream.cxx:32:
F:\opencv\3rdparty\libtiff\tiff.h:68: error: 'TIFF_INT8_T' does not name

Basically, there is no reason to bundle internal copy of libtiff if
the fix is already present in a more recent copy.

Thanks for the hint!

Oh, my God, there are two "tiff.h" in my working copy.

One is "F:\opencv\3rdparty\libtiff\tiff.h" I copied from the libtiff code, hence this is not belong to SVN code, the other is "F:\opencv\3rdparty\include\tiff.h". So the first one should be removed.

After that, I can successfully bulild OpenCV library.

Sorry for the noise to the mail list.

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
trial. Simplify your report design, integration and deployment - and focus on
what you do best, core application coding. Discover what's new with
Crystal Reports now. http://p.sf.net/sfu/bobj-july
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

OpenCV rev 2336 build failed in MinGW 4.4

2009-11-30T05:57:54Z

I try to build the latest SVN code under TDM-GCC 4.4.1 and Cmake 2.8.0

There is the build error, it seems the type is not declared it seems
this is introduces in the commit 2336.

Building CXX object 3rdparty/libtiff/CMakeFiles/libtiff.dir/tif_stream.obj
cd F:\test1\3rdparty\libtiff && D:\MinGW\bin\g++.exe -DHAVE_CONFIG_H
-Wall -ffunction-sections -g -O0 -DDEBUG -D_DEBUG -IF:\opencv\.
-IF:\test1 -IF:\opencv\include -IF:\opencv\include\opencv
-IF:\opencv\3rdparty\libtiff -IF:\opencv\3rdparty\libtiff\..\include
-o CMakeFiles\libtiff.dir\tif_stream.obj -c
F:\opencv\3rdparty\libtiff\tif_stream.cxx
In file included from F:\opencv\3rdparty\libtiff\tiffio.h:33,
from F:\opencv\3rdparty\libtiff\tiffiop.h:53,
from F:\opencv\3rdparty\libtiff\tif_stream.cxx:32:
F:\opencv\3rdparty\libtiff\tiff.h:68: error: 'TIFF_INT8_T' does not name
a type
F:\opencv\3rdparty\libtiff\tiff.h:69: error: 'TIFF_UINT8_T' does not
name a type
F:\opencv\3rdparty\libtiff\tiff.h:71: error: 'TIFF_INT16_T' does not
name a type
F:\opencv\3rdparty\libtiff\tiff.h:72: error: 'TIFF_UINT16_T' does not
name a type
F:\opencv\3rdparty\libtiff\tiff.h:74: error: 'TIFF_INT32_T' does not
name a type
F:\opencv\3rdparty\libtiff\tiff.h:75: error: 'TIFF_UINT32_T' does not
name a type
F:\opencv\3rdparty\libtiff\tiff.h:77: error: 'TIFF_INT64_T' does not
name a type
F:\opencv\3rdparty\libtiff\tiff.h:78: error: 'TIFF_UINT64_T' does not
name a type
F:\opencv\3rdparty\libtiff\tiff.h:94: error: 'uint16' does not name a type
F:\opencv\3rdparty\libtiff\tiff.h:95: error: 'uint16' does not name a type
F:\opencv\3rdparty\libtiff\tiff.h:98: error: 'uint16' does not name a type
F:\opencv\3rdparty\libtiff\tiff.h:99: error: 'uint16' does not name a type
F:\opencv\3rdparty\libtiff\tiff.h:100: error: 'uint32' does not name a type
F:\opencv\3rdparty\libtiff\tiff.h:103: error: 'uint16' does not name a type
F:\opencv\3rdparty\libtiff\tiff.h:104: error: 'uint16' does not name a type
F:\opencv\3rdparty\libtiff\tiff.h:105: error: 'uint16' does not name a type
F:\opencv\3rdparty\libtiff\tiff.h:106: error: 'uint16' does not name a type
F:\opencv\3rdparty\libtiff\tiff.h:107: error: 'uint64' does not name a type
In file included from F:\opencv\3rdparty\libtiff\tiffiop.h:53,
from F:\opencv\3rdparty\libtiff\tif_stream.cxx:32:
F:\opencv\3rdparty\libtiff\tiffio.h:67: error: 'TIFF_SSIZE_T' does not
name a type
F:\opencv\3rdparty\libtiff\tiffio.h:68: error: 'uint64' does not name a type
F:\opencv\3rdparty\libtiff\tiffio.h:71: error: 'uint32' does not name a type
F:\opencv\3rdparty\libtiff\tiffio.h:72: error: 'uint16' does not name a type
F:\opencv\3rdparty\libtiff\tiffio.h:73: error: 'uint16' does not name a type
F:\opencv\3rdparty\libtiff\tiffio.h:74: error: 'uint32' does not name a type
F:\opencv\3rdparty\libtiff\tiffio.h:75: error: 'tstrile_t' does not name
a type
F:\opencv\3rdparty\libtiff\tiffio.h:76: error: 'tstrile_t' does not name
a type
F:\opencv\3rdparty\libtiff\tiffio.h:77: error: 'tmsize_t' does not name
a type
F:\opencv\3rdparty\libtiff\tiffio.h:145: error: 'uint32' does not name a
type
F:\opencv\3rdparty\libtiff\tiffio.h:146: error: 'uint32' does not name a
type
F:\opencv\3rdparty\libtiff\tiffio.h:147: error: 'uint32' does not name a
type
F:\opencv\3rdparty\libtiff\tiffio.h:160: error: ISO C++ forbids
declaration of 'int32' with no type
F:\opencv\3rdparty\libtiff\tiffio.h:160: error: expected ';' before '*'
token
F:\opencv\3rdparty\libtiff\tiffio.h:161: error: ISO C++ forbids
declaration of 'int32' with no type
F:\opencv\3rdparty\libtiff\tiffio.h:161: error: expected ';' before '*'
token
F:\opencv\3rdparty\libtiff\tiffio.h:162: error: ISO C++ forbids
declaration of 'int32' with no type
F:\opencv\3rdparty\libtiff\tiffio.h:162: error: expected ';' before '*'
token
F:\opencv\3rdparty\libtiff\tiffio.h:191: error: 'uint32' has not been
declared
F:\opencv\3rdparty\libtiff\tiffio.h:191: error: 'uint32' has not been
declared
F:\opencv\3rdparty\libtiff\tiffio.h:191: error: 'uint32' has not been
declared
F:\opencv\3rdparty\libtiff\tiffio.h:191: error: 'uint32' has not been
declared
F:\opencv\3rdparty\libtiff\tiffio.h:191: error: 'uint32' has not been
declared
F:\opencv\3rdparty\libtiff\tiffio.h:191: error: 'int32' has not been
declared
F:\opencv\3rdparty\libtiff\tiffio.h:191: error: 'int32' has not been
declared
F:\opencv\3rdparty\libtiff\tiffio.h:194: error: 'uint32' has not been
declared
F:\opencv\3rdparty\libtiff\tiffio.h:194: error: 'uint32' has not been
declared
F:\opencv\3rdparty\libtiff\tiffio.h:194: error: 'uint32' has not been
declared
F:\opencv\3rdparty\libtiff\tiffio.h:194: error: 'uint32' has not been
declared
F:\opencv\3rdparty\libtiff\tiffio.h:194: error: 'uint32' has not been
declared
F:\opencv\3rdparty\libtiff\tiffio.h:194: error: 'int32' has not been
declared
F:\opencv\3rdparty\libtiff\tiffio.h:194: error: 'int32' has not been
declared
Process terminated with status 2 (3 minutes, 56 seconds)
50 errors, 29 warnings

Thanks.

Asmwarrior

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
trial. Simplify your report design, integration and deployment - and focus on
what you do best, core application coding. Discover what's new with
Crystal Reports now. http://p.sf.net/sfu/bobj-july
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

proforma - Probabilistic Feature-based On-line Rapid Model Acquisition

2009-11-28T02:46:38Z

Hello all,
i think this link about PRoforma with video+paper is quite interesting:

http://mi.eng.cam.ac.uk/~qp202/

Cheers,
Stefano

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
trial. Simplify your report design, integration and deployment - and focus on
what you do best, core application coding. Discover what's new with
Crystal Reports now. http://p.sf.net/sfu/bobj-july
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

blas opencl

2009-11-25T14:58:26Z

Hello Vadim,
is there any plan or interest about opencl support in opencv?
I see that there is an initial blas opencl code on sourceforge[¹].

[¹]http://sourceforge.net/projects/openclblas/

Cheers,
Stefano

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
trial. Simplify your report design, integration and deployment - and focus on
what you do best, core application coding. Discover what's new with
Crystal Reports now. http://p.sf.net/sfu/bobj-july
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

Re: libraries compiled with OMP on face recognition

2009-11-25T06:45:21Z

Vadim Pisarevsky wrote:

Hello,

sorry for a long delay. We also get problems with facedetection built with
OpenMP support. We are now studying the problem.

Regards,
Vadim

Thank you.
I wonder if the reason is always the corruption of parameter "minNeighbors" in void detectMultiScale(...).

Re: StarDetector code error

2009-11-25T05:19:32Z

Hello,

The current code is correct. If the row width is not divisible by 4, you will need to process the tail using a scalar code. That is actually implemented in the code (note that "x" is only set once in line 214).

Regards,

Vadim

On Wed, Nov 25, 2009 at 4:05 PM, Pierre Moulon <pmoulon@...> wrote:

Hello,
I want to notice you that in src/StarDetector I have spotted an error :
The loop to compute BiLevel responses is done two time IF you enable SSE2 support.

Notice that in line 261 you have put a #ENDIF. But it must be a #else ... and the else must finish in line 289 (after the '}' ).

So when you are in SS2 mode you compute two time the Bi-level response !!
I do not have tested yet. But i'm sure that some code are duplicated if you use SSE2 define...

Thanks to consider my opinion.

We have to test it before do any commit.

Best Regards,
Pierre M

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
trial. Simplify your report design, integration and deployment - and focus on
what you do best, core application coding. Discover what's new with
Crystal Reports now. http://p.sf.net/sfu/bobj-july
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel