OpenCV VideoCapture Failure from QThread

SGaist · 21 Jun 2019, 22:58

Hi,

Are you subclassing QThread ?

rtavakko · 22 Jun 2019, 01:42

@SGaist Yes I am. I've been told this is not the correct way to use QThreads but other say its just a matter of choice. Should I try passing a QObject to a thread instead?

Its worth mentioning that the problem is narrowed down to the .read() method of VideoCapture. Loading doesn't cause a crash. I tried using a lock and mutex right before reading a frame which seems to avoid a crash but the .read() method still hangs. Using mutex alone (locked before read, unlocked after) causes a crash as before.

JKSH · 22 Jun 2019, 04:09

@rtavakko said in OpenCV VideoCapture Failure from QThread:

I've been told this is not the correct way to use QThreads but other say its just a matter of choice.

Subclassing QThreads is a valid approach, but you must do it correctly.

Anyway, it sounds like you are calling VideoCapture methods from multiple threads. Is that class thread-safe? Do you know if you are allowed to call its methods from multiple threads?

rtavakko · 22 Jun 2019, 06:04

@JKSH I override the run function of the subclassed QThread object. I only call VideoCapture methods from a single QThread. These aren't thread-safe from what I've read online but even so I shouldn't have an issue if accessing them from one thread only unless there's something going on in the background that I don't know.

The issue is that if I move the VideoCaptures to the main thread, the issue is fixed but I get crashes when they are part of the QThread.

JKSH · 22 Jun 2019, 19:25

@rtavakko said in OpenCV VideoCapture Failure from QThread:

I only call VideoCapture methods from a single QThread.

Important note: Just because you call something from within a QThread method, that does not mean the code runs in the related thread!

Earlier, you said

The buttons end up calling this function in the QThread:

Well, Test::load() is running in the main thread, not your other thread.

If you want to verify this yourself, add qDebug() << QThread::currentThread() into the top of main(), the top of your run() function, and the top of Test::load(). This will show you which thread is running each function.

Read the QThread documentation carefully: https://doc.qt.io/qt-5/qthread.html. It says, "...a developer who wishes to invoke slots in the new thread must use the worker-object approach; new slots should not be implemented directly into a subclassed QThread.

When you reimplemented QThread::run(), you used an infinite loop (while (true)). This means your thread does not have an event loop, which means you cannot "trigger" anything in that thread.

rtavakko · J JKSH 22 Jun 2019, 06:04

@JKSH Thanks for the explanation. When a GUI button is pressed the thread is QThread(0x1922e9b0) and when the Test::load() is called, the thread is Test(0x197ded38) which is a bit confusing in terms of the names but looks like they are separate. I'm going to switch to what the documentation outlines and get back with the results.
On this note, how do I re-implement the event loop or make the thread continuously run without overriding the run() functions?

Christian Ehrlicher · wrote on 22 Jun 2019, 21:24

@rtavakko said in OpenCV VideoCapture Failure from QThread:

On this note, how do I re-implement the event loop or make the thread continuously run without overriding the run() functions?

See the docs: https://doc.qt.io/qt-5/qthread.html#run

"... The default implementation simply calls exec()."

rtavakko · 29 Jun 2019, 03:13

@JKSH I went back and noticed I had the qDebug() statement in the wrong place. You were correct about which thread a called method runs in. If the main thread calls a method in another thread, that method still runs in the main thread as the documentation says.

The issue was related to thread-safety as you suggested. Its also important to make sure we are not reading from a VideoCapture device that is in the process of loading a new video.

Here are my 'read' / 'load' methods for anyone having issues with this:

void Test::load(const int &vid, const int &file)
{
    QMutexLocker lock(&mutex);

    loading[vid] = true;

    switch(file)
    {
        case 0:
            loadFlag[vid] = video[vid].open(QDir::currentPath().toStdString()+"/vid0.mp4",cv::CAP_ANY);
            break;
        case 1:
            loadFlag[vid] = video[vid].open(QDir::currentPath().toStdString()+"/vid1.mp4",cv::CAP_ANY);
            break;
        case 2:
            loadFlag[vid] = video[vid].open(QDir::currentPath().toStdString()+"/vid2.mp4",cv::CAP_ANY);
            break;
    }
    loading[vid] = false;
}

bool Test::read(const int &vid)
{
    loadFlag[vid] = video[vid].isOpened();

    if(!loadFlag)
    {
        video[vid].release();
        return false;
    }
    else
    {
        QMutexLocker locker(&mutex);
        readFlag[vid] = video[vid].read(readMat[vid]);

        if(!readFlag[vid])
        {
            video[vid].release();
            return readFlag[vid];
        }
        else
        {
            if(!readMat[vid].empty())
            {
                cv::cvtColor(readMat[vid],readMat[vid],cv::COLOR_BGR2RGB);
                imgOut[vid] = QImage(readMat[vid].data, readMat[vid].cols, readMat[vid].rows, QImage::Format_RGB888);
                return true;
            }
        }
    }
}

Also the run method in the QThread object:

void Test::run()
{
    while(true)
    {
        for(int i = 0; i < 2; i++)
        {
            if(!loading[i])
            {
                bool tmp = this->read(i);
                if(tmp)
                {
                    emit processedImage(i,imgOut[i]);
                }
                else
                {
                    //SH*T HAPPENED
                }
            }
        }
    }
}

I'm still a bit confused about where exec() would come in, the documentation is a little hazy to me. Any SAFE threads you guys could reference that clears up how I need to implement run() and exec() together?

Thanks a lot for all of your help with this. Much appreciated!

JKSH · 2 Jul 2019, 18:35

@rtavakko said in OpenCV VideoCapture Failure from QThread:

I'm still a bit confused about where exec() would come in, the documentation is a little hazy to me. Any SAFE threads you guys could reference that clears up how I need to implement run() and exec() together?

The default implementation of QThread::run() is:

void QThread::run() {
    this->exec(); // Starts an event loop in the thread
}

If you want to reimplement run(), you have 2 options:

// Option 1: Run an event loop
void MyEventLoopThread::run() {
    /* Add your custom initialization code here */

    this->exec(); // Starts an event loop in the thread
}

// Option 2: Don't run an event loop; manually write an infinite loop
void MyPollingLoopThread::run() {
    /* Add your custom initialization code here */

    while ( !this->isInterruptionRequested() ) // Starts an infinite loop in the thread
    {
        /* Add your custom looping code here */
    }

Notice that Option #1 is just like your main thread:

int main(int argc, char **argv) {
    QApplication app(argc, argv);

    /* Add your custom initialization code here */

    app.exec(); // Starts an event loop in the main thread
}

Finally, Option #1 and Option #2 cannot be combined. Your thread can have either an event loop (exec()) OR a polling loop (while (true)). It cannot have both.

If you choose Option #1, your thread can receive signals.
If you choose Option #2, your thread cannot receive signals.

Does this all make sense?

rtavakko · 3 Jul 2019, 00:04

@JKSH Thanks a lot for taking the time to explain. The concept makes sense. After this->exec() is run with the event loop approach, the thread object is listed for deletion and its run() method cannot be called again. Is that correct? If so, then the thread object has to be created every time?

JKSH · 6 Jul 2019, 12:50

You're welcome.

@rtavakko said in OpenCV VideoCapture Failure from QThread:

After this->exec() is run with the event loop approach, the thread object is listed for deletion and its run() method cannot be called again. Is that correct? If so, then the thread object has to be created every time?

Nope, exec() does not mark anything for deletion.

You can certainly start() and quit() a thread with an event loop multiple times. (Note: Quit does nothing if your thread has no event loop)

rtavakko · J JKSH 3 Jul 2019, 00:04

@JKSH That makes sense. I'll give that a try. Thanks again!

rtavakko · 11 Sept 2019, 20:28

Hey guys,

I'm still seeing this same crash happen when I load new videos into an OpenCV VideoCapture object. It is a lot less frequent, it happens when I test the GUI loading files quickly one after the other.

But what is a bit confusing is that I don't get any details about what goes wrong exactly (reading a few posts with this message I understand this is related to an invalid pointer being accessed somewhere). I've tried puting try / catch statements various places in my code to get something.

What I get from QT test when running the same test automatically and the crash happens is something like this:

A crash occurred in C:\Users\Roham\Desktop\Lyrha\Modules\build-MorphTest-Desktop_Qt_5_12_0_MinGW_64_bit-Release\release\tst_morphtesttest.exe.
Function time: 33587ms Total time: 34304ms

Exception address: 0x0000000061b8afac
Exception code   : 0xc0000005
Nearby symbol    : ZN13QColorProfile8fromSRgbEv

Stack:
#  1: ZN5QTest15toPrettyUnicodeEPKti() - 0x000000006edc2750
#  2: UnhandledExceptionFilter() - 0x00007fff41b0f490
#  3: memset() - 0x00007fff447a2fc0
#  4: _C_specific_handler() - 0x00007fff4478c640
#  5: _chkstk() - 0x00007fff447a10e0
#  6: RtlRaiseException() - 0x00007fff44769ef0
#  7: KiUserExceptionDispatcher() - 0x00007fff4479fe40
#  8: ZN13QColorProfile8fromSRgbEv() - 0x0000000061b1e470
#  9: ZN13QVulkanWindow11qt_metacallEN11QMetaObject4CallEiPPv() - 0x0000000061d1c190
# 10: ZN13QColorProfile8fromSRgbEv() - 0x0000000061b1e470
# 11: ZrsR11QDataStreamR8QPolygon() - 0x0000000061c29620
# 12: ZN18QRasterPaintEngine4clipERK11QVectorPathN2Qt13ClipOperationE() - 0x0000000061bd93d0
# 13: ZN8QPainter9drawImageERK6QRectFRK6QImageS2_6QFlagsIN2Qt19ImageConversionFlagEE() - 0x0000000061bf2660
# 14: Unable to obtain symbol
# 15: ZN7QWidget5eventEP6QEvent() - 0x0000000001425ba0
# 16: ZN19QApplicationPrivate13notify_helperEP7QObjectP6QEvent() - 0x00000000013e7b70
# 17: ZN12QApplication6notifyEP7QObjectP6QEvent() - 0x00000000013ee990
# 18: ZN16QCoreApplication20sendSpontaneousEventEP7QObjectP6QEvent() - 0x0000000068a674f0
# 19: ZN14QWidgetPrivate14sendPaintEventERK7QRegion() - 0x000000000141e0f0
# 20: ZN14QWidgetPrivate10drawWidgetEP12QPaintDeviceRK7QRegionRK6QPointiP8QPainterP19QWidgetBackingStore() - 0x000000000141e150
# 21: ZN14QWidgetPrivate22paintSiblingsRecursiveEP12QPaintDeviceRK5QListIP7QObjectEiRK7QRegionRK6QPointiP8QPainterP19QWidgetBackingStore() - 0x000000000141f220
# 22: ZN14QWidgetPrivate22paintSiblingsRecursiveEP12QPaintDeviceRK5QListIP7QObjectEiRK7QRegionRK6QPointiP8QPainterP19QWidgetBackingStore() - 0x000000000141f220
# 23: ZN14QWidgetPrivate22paintSiblingsRecursiveEP12QPaintDeviceRK5QListIP7QObjectEiRK7QRegionRK6QPointiP8QPainterP19QWidgetBackingStore() - 0x000000000141f220
# 24: ZN14QWidgetPrivate22paintSiblingsRecursiveEP12QPaintDeviceRK5QListIP7QObjectEiRK7QRegionRK6QPointiP8QPainterP19QWidgetBackingStore() - 0x000000000141f220
# 25: ZN14QWidgetPrivate10drawWidgetEP12QPaintDeviceRK7QRegionRK6QPointiP8QPainterP19QWidgetBackingStore() - 0x000000000141e150
# 26: ZN14QWidgetPrivate22paintSiblingsRecursiveEP12QPaintDeviceRK5QListIP7QObjectEiRK7QRegionRK6QPointiP8QPainterP19QWidgetBackingStore() - 0x000000000141f220
# 27: ZN14QWidgetPrivate10drawWidgetEP12QPaintDeviceRK7QRegionRK6QPointiP8QPainterP19QWidgetBackingStore() - 0x000000000141e150
# 28: ZN14QWidgetPrivate11repaint_sysERK7QRegion() - 0x00000000013f41f0
# 29: ZN14QWidgetPrivate11repaint_sysERK7QRegion() - 0x00000000013f41f0
# 30: ZN14QWidgetPrivate16syncBackingStoreEv() - 0x000000000140d160
# 31: ZN7QWidget5eventEP6QEvent() - 0x0000000001425ba0
# 32: ZN11QMainWindow5eventEP6QEvent() - 0x0000000001513a50
# 33: ZN19QApplicationPrivate13notify_helperEP7QObjectP6QEvent() - 0x00000000013e7b70
# 34: ZN12QApplication6notifyEP7QObjectP6QEvent() - 0x00000000013ee990
# 35: ZN16QCoreApplication15notifyInternal2EP7QObjectP6QEvent() - 0x0000000068a66cc0
# 36: ZN23QCoreApplicationPrivate16sendPostedEventsEP7QObjectiP11QThreadData() - 0x0000000068a6d370
# 37: qt_plugin_instance() - 0x000000006a8f6eb0
# 38: ZN28QEventDispatcherWin32Private14sendTimerEventEi() - 0x0000000068ac1680
# 39: CallWindowProcW() - 0x00007fff427c6030
# 40: DispatchMessageW() - 0x00007fff427c5bf0
# 41: ZN21QEventDispatcherWin3213processEventsE6QFlagsIN10QEventLoop17ProcessEventsFlagEE() - 0x0000000068ac0b60
# 42: qt_plugin_instance() - 0x000000006a8f6eb0
# 43: ZN16QCoreApplication13processEventsE6QFlagsIN10QEventLoop17ProcessEventsFlagEEi() - 0x0000000068a67250
# 44: ZN5QTest5qWaitEi() - 0x0000000068ab7a20
# 45: Unable to obtain symbol
# 46: Unable to obtain symbol
# 47: Unable to obtain symbol
# 48: Unable to obtain symbol
# 49: Unable to obtain symbol
# 50: Unable to obtain symbol
# 51: ZNK11QMetaMethod6invokeEP7QObjectN2Qt14ConnectionTypeE22QGenericReturnArgument16QGenericArgumentS5_S5_S5_S5_S5_S5_S5_S5_S5_() - 0x0000000068a74760
# 52: ZN5QTest5qInitEP7QObjectiPPc() - 0x000000006edc7590
# 53: ZN5QTest5qInitEP7QObjectiPPc() - 0x000000006edc7590
# 54: ZN5QTest5qInitEP7QObjectiPPc() - 0x000000006edc7590
# 55: ZN5QTest4qRunEv() - 0x000000006edca280
# 56: ZN5QTest5qExecEP7QObjectiPPc() - 0x000000006edca670
# 57: Unable to obtain symbol
# 58: Unable to obtain symbol
# 59: Unable to obtain symbol
# 60: BaseThreadInitThunk() - 0x00007fff44077bc0
# 61: RtlUserThreadStart() - 0x00007fff4476ceb0

The program has unexpectedly finished.
The process was ended forcefully.

Any tips on how I could try and narrow down the issue?

There are mentions of threads and QPainter but I'm sure the issue is somehow related to loading a new file but I'm not sure how to get more information.

Cheers!

JKSH · 10 Nov 2019, 14:29

@rtavakko Your stack trace says that the exception was thrown in QColorProfile::fromSRgb().

This is a private class and method that was removed in Qt 5.13 -- perhaps you could try to upgrade to Qt 5.13 (or even Qt 5.14 beta) and see if the crash still occurs?

rtavakko · 12 Nov 2019, 01:00

@JKSH Thank for the reply! Aaah so 'Nearby Symbol' is where it occured and 'Exception Code' is what the error was. For future debugging is there a way I could get the stack trace when running the project normally? Currently the only way I could get it was by using QT unit test to start the GUI application.

I will try upgrading to QT 5.13. Any idea what could be inviting QColorProfile::fromSRgb() to the party?

I get this warning a few times everytime I start the project:

libpng warning: iCCP: known incorrect sRGB profile

I looked around online and people said it wasn't really a big deal but it seems it is related to the method that caused the crash.

Cheers!

rtavakko · wrote on 11 Nov 2019, 17:10

Update on this. I got the same crash in Qt 5.13.2 (Mingw 7.3.0 64-bit). I got a new crash in Qt 5.14.0 (same test conditions and crash happens the same way as before):

Exception address: 0x0000000061bc697c
Exception code   : 0xc0000005
Nearby symbol    : ZN12QColorTrcLut17fromTransferTableERK19QColorTransferTable

Stack:
#  1: ZN5QTest15toPrettyUnicodeEPKti() - 0x000000006edc27a0
#  2: UnhandledExceptionFilter() - 0x00007ff9f924f490
#  3: memset() - 0x00007ff9fbc82fc0
#  4: _C_specific_handler() - 0x00007ff9fbc6c640
#  5: _chkstk() - 0x00007ff9fbc810e0
#  6: RtlRaiseException() - 0x00007ff9fbc49ef0
#  7: KiUserExceptionDispatcher() - 0x00007ff9fbc7fe40
#  8: ZN12QColorTrcLut17fromTransferTableERK19QColorTransferTable() - 0x0000000061b4b770
#  9: ZN13QVulkanWindow11qt_metacallEN11QMetaObject4CallEiPPv() - 0x0000000061dbb980
# 10: ZN12QColorTrcLut17fromTransferTableERK19QColorTransferTable() - 0x0000000061b4b770
# 11: ZrsR11QDataStreamR8QPolygon() - 0x0000000061c60140
# 12: ZN18QRasterPaintEngine4clipERK11QVectorPathN2Qt13ClipOperationE() - 0x0000000061c0ff20
# 13: ZN8QPainter9drawImageERK6QRectFRK6QImageS2_6QFlagsIN2Qt19ImageConversionFlagEE() - 0x0000000061c29840
# 14: Unable to obtain symbol
# 15: ZN7QWidget5eventEP6QEvent() - 0x00000000012452f0
# 16: ZN19QApplicationPrivate13notify_helperEP7QObjectP6QEvent() - 0x0000000001207ae0
# 17: ZN12QApplication6notifyEP7QObjectP6QEvent() - 0x000000000120e7f0
# 18: ZN16QCoreApplication20sendSpontaneousEventEP7QObjectP6QEvent() - 0x0000000068a7dc40
# 19: ZN14QWidgetPrivate14sendPaintEventERK7QRegion() - 0x000000000123d310
# 20: ZN14QWidgetPrivate10drawWidgetEP12QPaintDeviceRK7QRegionRK6QPoint6QFlagsINS_14DrawWidgetFlagEEP8QPainterP21QWidgetRepaintManager() - 0x000000000123d370
# 21: ZN14QWidgetPrivate22paintSiblingsRecursiveEP12QPaintDeviceRK5QListIP7QObjectEiRK7QRegionRK6QPoint6QFlagsINS_14DrawWidgetFlagEEP8QPainterP21QWidgetRepaintManager() - 0x000000000123eb50
# 22: ZN14QWidgetPrivate22paintSiblingsRecursiveEP12QPaintDeviceRK5QListIP7QObjectEiRK7QRegionRK6QPoint6QFlagsINS_14DrawWidgetFlagEEP8QPainterP21QWidgetRepaintManager() - 0x000000000123eb50
# 23: ZN14QWidgetPrivate22paintSiblingsRecursiveEP12QPaintDeviceRK5QListIP7QObjectEiRK7QRegionRK6QPoint6QFlagsINS_14DrawWidgetFlagEEP8QPainterP21QWidgetRepaintManager() - 0x000000000123eb50
# 24: ZN14QWidgetPrivate22paintSiblingsRecursiveEP12QPaintDeviceRK5QListIP7QObjectEiRK7QRegionRK6QPoint6QFlagsINS_14DrawWidgetFlagEEP8QPainterP21QWidgetRepaintManager() - 0x000000000123eb50
# 25: ZN14QWidgetPrivate10drawWidgetEP12QPaintDeviceRK7QRegionRK6QPoint6QFlagsINS_14DrawWidgetFlagEEP8QPainterP21QWidgetRepaintManager() - 0x000000000123d370
# 26: ZN14QWidgetPrivate22paintSiblingsRecursiveEP12QPaintDeviceRK5QListIP7QObjectEiRK7QRegionRK6QPoint6QFlagsINS_14DrawWidgetFlagEEP8QPainterP21QWidgetRepaintManager() - 0x000000000123eb50
# 27: ZN14QWidgetPrivate10drawWidgetEP12QPaintDeviceRK7QRegionRK6QPoint6QFlagsINS_14DrawWidgetFlagEEP8QPainterP21QWidgetRepaintManager() - 0x000000000123d370
# 28: ZNK14QWidgetPrivate24shouldDiscardSyncRequestEv() - 0x00000000012133a0
# 29: ZNK14QWidgetPrivate24shouldDiscardSyncRequestEv() - 0x00000000012133a0
# 30: ZN14QWidgetPrivate16syncBackingStoreEv() - 0x000000000123e9e0
# 31: ZN7QWidget5eventEP6QEvent() - 0x00000000012452f0
# 32: ZN11QMainWindow5eventEP6QEvent() - 0x0000000001338e70
# 33: ZN19QApplicationPrivate13notify_helperEP7QObjectP6QEvent() - 0x0000000001207ae0
# 34: ZN12QApplication6notifyEP7QObjectP6QEvent() - 0x000000000120e7f0
# 35: ZN16QCoreApplication15notifyInternal2EP7QObjectP6QEvent() - 0x0000000068a7d410
# 36: ZN23QCoreApplicationPrivate16sendPostedEventsEP7QObjectiP11QThreadData() - 0x0000000068a83440
# 37: qt_plugin_instance() - 0x000000006a8fbe40
# 38: ZN21QEventDispatcherWin3213processEventsE6QFlagsIN10QEventLoop17ProcessEventsFlagEE() - 0x0000000068ad8cb0
# 39: qt_plugin_instance() - 0x000000006a8fbe40
# 40: ZN16QCoreApplication13processEventsE6QFlagsIN10QEventLoop17ProcessEventsFlagEEi() - 0x0000000068a7d9a0
# 41: ZN5QTest5qWaitEi() - 0x0000000068acfbe0
# 42: Unable to obtain symbol
# 43: Unable to obtain symbol
# 44: Unable to obtain symbol
# 45: Unable to obtain symbol
# 46: Unable to obtain symbol
# 47: Unable to obtain symbol
# 48: ZNK11QMetaMethod6invokeEP7QObjectN2Qt14ConnectionTypeE22QGenericReturnArgument16QGenericArgumentS5_S5_S5_S5_S5_S5_S5_S5_S5_() - 0x0000000068a8b460
# 49: ZN5QTest5qInitEP7QObjectiPPc() - 0x000000006edc88d0
# 50: ZN5QTest5qInitEP7QObjectiPPc() - 0x000000006edc88d0
# 51: ZN5QTest5qInitEP7QObjectiPPc() - 0x000000006edc88d0
# 52: ZN5QTest4qRunEv() - 0x000000006edcb5c0
# 53: ZN5QTest5qExecEP7QObjectiPPc() - 0x000000006edcb9b0
# 54: Unable to obtain symbol
# 55: Unable to obtain symbol
# 56: Unable to obtain symbol
# 57: BaseThreadInitThunk() - 0x00007ff9faab7bc0
# 58: RtlUserThreadStart() - 0x00007ff9fbc4ceb0

JKSH · 12 Nov 2019, 03:01

@rtavakko said in OpenCV VideoCapture Failure from QThread:

Aaah so 'Nearby Symbol' is where it occured and 'Exception Code' is what the error was.

Yep. And the stack trace shows the calls leading up to the crash, in reverse order.

In Qt 5.12 you had: QVulkanWindow::qt_metacall(...) -> QColorProfile::fromSRgb() -> crash
In Qt 5.14 you had: QVulkanWindow::qt_metacall(...) -> QColorTrcLut::fromTransferTable() -> crash

Normally, the function where the crash occured will be item #1 on the stack trace because it's the last one called. However, since you're running inside a test framework, it was able to "catch" the crash and do some more work to present the crash to you (starting from UserExceptionDispatcher())

Any idea what could be inviting QColorProfile::fromSRgb() to the party?

Judging by your stack trace, your system is in the middle of decoding or rendering the video when the crash occurs.

For future debugging is there a way I could get the stack trace when running the project normally? Currently the only way I could get it was by using QT unit test to start the GUI application.

You can just run your debugger without using Qt Test: https://doc.qt.io/qtcreator/creator-debugging.html It is useful for locating "simple" coding errors like trying to read beyond the last element of an array.

However, in your situation, you get a memory access violation error (https://stackoverflow.com/questions/5303524/what-exactly-is-the-scope-of-access-violation-0xc0000005 ) when loading many files quickly but doesn't crash when loading slowly, is that right? In this situation, the debugger is less useful. I'm guessing that when you have many threads loading at the same time, one of the threads could be corrupting the memory used by another thread and this corruption is detected during the colour lookup (https://code.woboq.org/qt5/qtbase/src/gui/painting/qcolortrclut.cpp.html#_ZN12QColorTrcLut17fromTransferTableERK19QColorTransferTable ). The corruption was probably not caused by the colour lookup function itself.

I don't know of a good way to debug threading issues in Windows. If you can build your project in Linux, Valgrind might help.

libpng warning: iCCP: known incorrect sRGB profile
I looked around online and people said it wasn't really a big deal but it seems it is related to the method that caused the crash.

I don't think libpng is used to load videos. Is your program loading any images?

rtavakko · 12 Nov 2019, 05:41

@JKSH Thanks for the detailed information, it'll definitely help with future bugs like this.

So far, it seems to crash more frequently when I load files quickly but I also tried slowing down the test and it still happened with no real pattern.

I have QMutexlockers in key areas where I load and read videos but after doing some more testing, it seems the crash happens only when I emit a signal containing an OpenCV matrix data pointer. I use this signal to display image results using a subclassed widget. If I don't emit the signal and use an OpenCV window to show an image, everything runs well no matter how fast I load videos. So I think there's something wrong in my handling of the sent pointer.

//Show current image data
void Display::updateImage(const uchar *imageData, int width, int height, int step)
{
    if(imageData != nullptr)
    {
        displayImage = QImage(imageData, width, height, step, QImage::Format_RGBA8888);
        update();
    }
}

//Paint event for display object
void Display::paintEvent(QPaintEvent*)
{
    QPainter painter(this);

    painter.setRenderHint(QPainter::SmoothPixmapTransform,true);
    painter.setPen(QPen(Qt::red));

    painter.drawImage(rect(),displayImage); //Resize the image based on widget size and display it
}

I don't see what in here could be causing a crash specially given that I use a const pointer. Is there a way I could catch the exception when it happens if I can't get to the bottom of the issue? So far try / catch statements haven't helped.

Cheers!

JKSH · 11 Dec 2019, 05:43

@rtavakko said in OpenCV VideoCapture Failure from QThread:

it seems the crash happens only when I emit a signal containing an OpenCV matrix data pointer

I don't have much experience with OpenCV, but you might need to be extra careful when passing cv::mat across different threads: https://stackoverflow.com/questions/10657841/is-cvmat-thread-safe-atomic-assignment-refcounting

I don't see what in here could be causing a crash specially given that I use a const pointer.

The code you posted does not show how you use cv::mat or how you emit your signal.

Is there a way I could catch the exception when it happens if I can't get to the bottom of the issue?

Not really. Multithreading-related memory problems cause damage long before the access violation exception is generated.

SGaist · wrote on 12 Nov 2019, 07:28

Hi,

@rtavakko said in OpenCV VideoCapture Failure from QThread:

void Display::updateImage(const uchar *imageData, int width, int height, int step)
{
if(imageData != nullptr)
{
displayImage = QImage(imageData, width, height, step, QImage::Format_RGBA8888);
update();
}
}

The constructor you are using does not make a copy of the data and you have to ensure that the lifetime of the buffer pointed to by imageData stays valid as long as the displayImage is.

From the looks of it, you should either call copy so that displayImage doesn't depend on the lifetime of imageData or create imageData once with the right dimensions and use memcpy to transfer the data. Re-create displayImage if the size changes at some point.