Issue with threading and opencv
-
Hello All,
I have written a program in opencv to fetch an image form camera and save in it folder.
There is another program which takes the program form the same folder and do some computation of the image.
I combined both the program and put the second logic in thread. Now the problem is what is figured is:
The main program tries to save the image in the same program and at the same time the thread tries to read different image form the same folder. So sometimes the image is saved but only half of the image is saved and other half contains no data. Also the thread breaks.
I tried using QMutex but the issue was the thread break if the main thread is already lock.
The summary of the problem is:
->How can i provide synchronization between the thread.
->How should i make the other thread to wait until the lock is not released.
-
Hi @Kira
a solution doesn't quit come to my mind, but I have to ask, why do save the image in the HD first ? Seems like an unnecessary step ?
-
@J.Hilk : Yes i agree with you its unnessary to save the image. In future we will store it internally in queue. But this is an interesting use case. I wanted to know the problem exactly so that i can avoid it in near future.
-
@Kira how do you save the file, also with Qt ? Than maybe QSaveFile will help
https://doc.qt.io/qt-5/qsavefile.html#details
That said, maybe using QFileSystemWatcher to monitor your saved file should also reduce the error frequency ?
-
@J.Hilk : I am using imwrite an opencv function to save the image.
Actually the problem arise after i start using thread. So i just want to know how the thread is causing the issue.
-
@Kira The thread writing the file should emit a signal when the file was written and closed. The thread consuming files should only read files which were already notified through that signal. This way you would avoid reading and writing same file from two different threads at the same time.
-
@jsulm : Just one question for clarification.
If i use two different mutex, will the one mutex will inform about other if i tried to lock ?
Or should i use same mutex where the conflicting situation is expected ?
Also i was going through opencv i got the following warning which was not seen before:
Warning:Premature end of JPEG file
-
@Kira said in Issue with threading and opecv:
If i use two different mutex, will the one mutex will inform about other if i tried to lock ?
No.
But with the solution I suggested there is no need at all for any mutexes. Just use signals/slots as I suggested with queued connection type."Warning:Premature end of JPEG file" - I guess it's because the file is damaged.
-
@jsulm : Thanks for highlighting the possibility of signal/slots.
I will have to redesign my approach for implementing the same.
Just couple of question:
-> If i queue signal and slot will it affect my overall program.
ie. Suppose i queue the signal/slots related to this program will it affect the other signal/slots
written in my program . For ex: delay in execution of the signal/slots of the main program.
-> How to make a mutex wait for the previous lock if we are not sure how much time it is going to take.
-
@Kira said in Issue with threading and opecv:
Suppose i queue the signal/slots related to this program will it affect the other signal/slots
written in my program . For ex: delay in execution of the signal/slots of the main program.Queued signals do not block anything - they are just in a queue. The only delay you will have is the slot when it is actually called - but that you will have anyway as you have to execute the code anyway, right?
"How to make a mutex wait for the previous lock if we are not sure how much time it is going to take" - I don't understand this question. A mutex either blocks the thread which wants to lock if the mutex is already locked, or you can continue immediately. So, you already have this "waiting" if the mutex is already locked.
-
@jsulm
Queued signals do not block anything - they are just in a queue. The only delay you will have is the slot when it is actually called - but that you will have anyway as you have to execute the code anyway, right? : YesThanks for the help cleared a lot of implementation doubts.
Just a small question. If i have a thread which acquires the lock and there is one more thread which requests for the locked resource. How will i tell the other thread wait until and unless the lock in the thread have not been unlocked.
We can currently use the wait time but is there any way to tell when create a notification when the lock is released and tell the other thread to wait for that specific time.And one more thing how to highlight specific lines which we are answering. Which you have did for my earlier questions. :)
-
@Kira said in Issue with threading and opecv:
How will i tell the other thread wait until
That's exactly what a mutex is doing: it blocks the thread which tries to lock a mutex which is already locked by other thread, as soon as the other thread releases the mutex the OS will wake up one of the blocked (waiting) threads. So, you do not have to tell the thread anything.
-
@jsulm : Hello i have implemented the save image function in a slot as suggested by you.
But the problem is that the image is not being saved. I would like to share the given code:signals:
void signalSaveImage(cv::Mat,String row,String column);
slots:
public slots:
void saveImage(Mat finalImage,String row, String column);
Slot Definition:void MainWindow::saveImage(Mat finalImage, String row, String column) { cout<<"Inside image save function"<<endl; cout<<"Row"<<row<<endl; cout<<"Column"<<column<<endl; String imagePath = "D:/testN/" String imageFormat = ".bmp"; try { cv::imwrite(imagePath+row+"_"+column+imageFormat,finalImage); } catch (cv::Exception& ex) { fprintf(stderr, "Exception in saving image: %s\n", ex.what()); } fprintf(stdout, "Saved PNG file with alpha data.\n"); cout<<"\n Entering inside save mutex"<<endl; cout<<"\n Completed image save function"<<endl; }Signal to slot connection:
connect(this,SIGNAL(signalSaveImage(cv::Mat,String,String)),this,SLOT(saveImage(Mat,String,String)),Qt::QueuedConnection);MainWindow.cpp
Emitting signal to save image:
emit signalSaveImage(image,to_string(row),to_string(column));But my problem is that my images are not being saved.
-
@Kira said in Issue with threading and opencv:
QueuedConnection
forces a copy of the arguments, The MetaSystem doesn't know the type cv::Mat -> can't be transmitted via QueuedConnection.
Check your console output you should have warning of unknown metatyps (declare and/or register)
-
@J.Hilk : I have registered it.
#include "mainwindow.h"
#include <QApplication>int main(int argc, char *argv[])
{
qRegisterMetaType< cv::Mat >("cv::Mat");
QApplication a(argc, argv);
MainWindow w;
w.show();return a.exec();}
-
the data goes most likely out of scope, as the copy constructor of cv::Mat does actually no coping
I had the same issues in my time with opencv.
ended up converting it to a Image before transferring it from the opencv thread to the normal one.
-
@J.Hilk : Just for clarification should i replace this part with cv::clone
void saveImage(Mat finalImage,String row, String column);
-
@J-Hilk : will try and also i tried to run the program in debug mode and getting the following error.
Debugger redirects to following line of file of ostream:
// MANIPULATORS
template<class _Elem,
class _Traits> inline
basic_ostream<_Elem, _Traits>&
__CLRCALL_OR_CDECL endl(basic_ostream<_Elem, _Traits>& _Ostr)
{ // insert newline and flush stream
_Ostr.put(_Ostr.widen('\n'));
_Ostr.flush();
return (_Ostr);
}
And the program quits.
Can u please explain what may be the relevant reason for error
-
@Kira Please post the stack trace after crash
-
@jsulm : I thought this is the stack trace after crash
-
-
@jsulm : This is the only thing which i see after crash with exception triggered.
Is there any other way to do so
-
@Kira In QtCreator if you start with debugger you will have a stack trace
-
@jsulm : Are you referring to this:
Please let me know in case of any issues
-
@Kira Yes. Now you need to go from top to bottom until you hit first source code file from your project and check it at the line mentioned in the stack trace.
-
@jsulm : I tried doing it but the error is expected to come after thread running >150 times which is making the debugging process very difficult.
I have also sample test case on github can you please go through it for any possible issues.
https://github.com/HackersSpirit/ImageProcessor
-
@Kira said in Issue with threading and opencv:
150 times
you can tell a breakpoint to be ignored for a certain amount of time
-
@Kira I actually suggested to go through the stack trace after crash from top to bottom. No break points needed for that.
-
@jsulm :
@Kira said in Issue with threading and opencv:@jsulm : Are you referring to this:
Please let me know in case of any issuesThis is the stack trace which i get posted earlier referring from line no 1. It does not point to line of code where program has a break.
But line 16 and 17 point to the line of the code.
Should i refer those line for particular cause of the program to break
Please do let me know if i getting something wrong
-
@Kira Check your code in that source code at line 16 and 17.
-
@jsulm :
Line 16 is pointing to following line in bold:
void ImageMapping::requestReWorkNew()
{
cout<<"Inside requestRework Newwwwwwwwww"<<endl;
const int sleep = 50;
int queueSize = queueImageName.size();
cout<<"Queue size Request Rework New: "<<queueSize<<endl;
if(queueImageName.size()!=0){
//Get the current image number from the list of images
mutex.lock();
_working = true;
_abort = false;
mutex.unlock();
cout<<"Emiting resume work"<<endl;
emit resumeWork();
}else{
cout<<"Signalling retrying"<<endl;emit retrying(); }}
This function is basically a slot triggered by following signal:
connect(imageMapping, SIGNAL(finished()), imageMapping, SLOT(requestReWorkNew()));
connect(imageMapping, SIGNAL(retrying()),imageMapping,SLOT(requestReWorkNew()));
-
@Kira Are you sure this is the line? The actual line is mentioned in the "Line" column in the stack trace. So, it would be 2129.
-
@Kira you can double click on the line (in QtCreator) of the stack trace and it will open the file at the correct line
-
@jsulm : Actually i have recreated the issue.the program i am referring now is from github. I have cross checked the line which I highlighted is the same. If u want i can repost the image
-
@J-Hilk @jsulm : guys have you gone through the code at github do you see any issue with the threading logic? :)
-
@J-Hilk @jsulm : Hello guys sorry for the late update i was working on the issue.
Would like to share the update.
I trimmed down the issue further by first commenting out all the cout statements as the logs which
was coming pointed to the ostream file of the cout.
So the call stack pointed out to the opencv file stating unable to save the image.
So i commented the save function to find any thing else was causing the issue. What i found that my program was still breaking and the call stack pointed to the thread function.
I placed a counter for the signal and slot of the thread and it was found that it executed 666 times approx before breaking.
-
@J-Hilk
@jsulm :Guys thanks for your support i finally figured where exactly the problem was:
There was issue with the logic which was used for implementation. As you can see the code on github.
Instead of requesting the task in the dowork function every time a signal is raised and an slot is called to execute the task.
This led to the calling of so many signals and slots within a given time that it caused the stack to overflow.
The issue will be resolved by removing the signal slot loop. This issue was top priority, a big learning curve for my future implementations in dealing with signal and slot mechanism.Thanks for your support.
Just one last question which i would like to be answered which triggered with this implementation.- Is there any limitation of how much slots can reside at a given time on a stack?
As per my observation at a given time 666 approx signal/slots calls were made after which the program crashed. - Is the slot entry is cleared or maintained on the stack after the execution and can be increase the size of stack or transfer it to heap?
Regards
- Is there any limitation of how much slots can reside at a given time on a stack?
-
@Kira 1. Sure but I guess it depends on many factors. You should not rely on it and avoid flooding the signal queue
2. Not sure I understand the question. Which slot entry do you mean? Slots are not put into the signal queue. And the stack is automatically cleared when a function/method (which a slot is) finishes, nothing Qt specific.
-
@jsulm said in Issue with threading and opencv:
@Kira 1. Sure but I guess it depends on many factors. You should not rely on it and avoid flooding the signal queue
2. Not sure I understand the question. Which slot entry do you mean? Slots are not put into the signal queue. And the stack is automatically cleared when a function/method (which a slot is) finishes, nothing Qt specific.- Sorry if not understood i will just try to clear :). Actually i my assumption due to the implementation logic there were so many signals generated that caused the stacks to overflow giving error at different points.
For example sometime i got error at saveImage function. while sometimes at calling the slot etc.
I still need to clarify my assumptions using different test cases but probably this should be the cause why the program crashed.
- Sorry if not understood i will just try to clear :). Actually i my assumption due to the implementation logic there were so many signals generated that caused the stacks to overflow giving error at different points.
-
@Kira said in Issue with threading and opencv:
- Sorry if not understood i will just try to clear :). Actually i my assumption due to the implementation logic there were so many signals generated that caused the stacks to overflow giving error at different points.
For example sometime i got error at saveImage function. while sometimes at calling the slot etc.
I still need to clarify my assumptions using different test cases but probably this should be the cause why the program crashed.
Hi, great that you managed to work it out thumbsUp
Sadly I'm quite a but under time pressure and was unable to take a closer look on your uploaded project :(
To your question.
The (default)stack limit is highly OS depended but In most cases, the max stack size of the main thread is significantly large, that you won't run into issues, even in highly recursive functions.
IIRC, than on an ordinary desktop PCs, the call stack for Windows defaults to 1mb, 8mb for Linux, and 8-64mb for OSXFor 2ndary threads, that can be quite different for macOS, that's reduced to 512 kbyte
- Sorry if not understood i will just try to clear :). Actually i my assumption due to the implementation logic there were so many signals generated that caused the stacks to overflow giving error at different points.
