Hi everyone,
It might be a stupid question but..why Visual Studio (2010) says there's a memory leak here? I'm using OpenCV 2.4.1 with TBB support.
int main()
{
Mat mask(480, 640, CV_8U, Scalar(255));
cv::erode(mask, mask, cv::Mat());
// Turn on memory leaks detection.
_CrtSetDbgFlag((_CRTDBG_LEAK_CHECK_DF) | _CrtSetDbgFlag(_CRTDBG_REPORT_FLAG));
return 0;
}
Regards,
Giuliano
Here is no memory leak.
As we know, OpenCV uses auto-initialization for TBB. So task dispatcher often will be destroyed when the application completes. This is the reason, that VS reports about memory leak. When you use local variable cv::Ptr, it will be destroyed as usual local variable.
You shouldn't care about when it will be destroyed (but it will) and use construction with cv::Ptr in usual application, like this example. I'm suggest to use more accurate tools for memory analyze in case of TBB, like this.
Asked: 2012-09-04 16:43:32 -0600
Seen: 1,998 times
Last updated: Sep 10 '12
Do you still have this leak if comment erode call?
Commenting the erode, the leak disappears. Same stuff using "dilate".
Thank you for response. Could you try to run erode in cycle and see in task manager is memory leaks or not. Now I'm on linux, in system monitor no memory leaks; checked with some tools, they aren't reported about leaks. May be it's windows specific?
If I run erode in a loop, the allocated memory does not increase. There's still only one memory leak. Checking where that memory is allocated, I can see some references to TBB code (starting from morph.cpp:1183).
I suppose, that in this example with erode threading isn't used, because tbb implementation doesn't support inplace calculation (source is a destination at the same time). Please, try to add cv::Ptr<tbb::task_scheduler_init>; tbb_init = new tbb::task_scheduler_init(); before erode call. Is now memory leak?
It seems the line of code fixes the issue! What's the explanation? I've never used TBB before.
Many thanks for testing, I've posted the answer.