IplImage* has memory leak, how to release memory ?

[pattacini]

@Tariq-Abuhashim commented on Fri Jun 24 2016

IplImage* cvImageL allocates the memory, and cvReleaseImage( &cvImageL ) is supposed to release it. Looks like it only deletes the pointer, but not the physical memory allocated.
This seems to be caused by cvLoadImage:
IplImage* cvImageL = cvLoadImage( left_img_file_name.c_str() );

==8172== LEAK SUMMARY:
==8172== definitely lost: 0 bytes in 0 blocks
==8172== indirectly lost: 0 bytes in 0 blocks
==8172== possibly lost: 0 bytes in 0 blocks
==8172== still reachable: 88 bytes in 3 blocks
==8172== suppressed: 0 bytes in 0 blocks
==8172== Rerun with --leak-check=full to see details of leaked memory

Any ideas on how to empty the allocated memory ?

---

@pattacini commented on Fri Jun 24 2016

It seems unrelated to stereo-vision, although stereo-vision still makes use of old OpenCV code.
Moving to Q&A...

[pattacini]

Hi @Tariq-Abuhashim

cvReleaseImage should actually free up the memory, if I'm not mistaken.
However, I'd warmly recommend you to rely on the C++ layer of OpenCV, thus replacing the use of *IplImage with Mat.

Therefore, look up the imread method, which should basically do what cvLoadImage is used to.

[lornat75]

It seems strange that cvReleaseImage() does not free all the previously allocated memory (e.g. by cvLoadImage( )). This would imply a huge memory leak in opencv.

BTW: the way I have found most convenient to pass a YARP Image to OpenCV is by creating a Mat object and initialize it from a IplImage pointer, i.e.:

ImageOf<PixelRgb> image;
// fill image content
Mat image_cv((IplImage *)image.getIplImage());
// now image_cv can be used in opencv

In my understanding this is quite efficient because image_cv shares storage allocated by image. Memory is therefore managed by YARP.

You just have to be careful because image_cv and image share their content.

Is there any better method? @vtikha

[gsaponaro]

Don't know about the leak, but just a small side comment on YARP + OpenCV. I basically use the same approach as @lornat75, with the addition of cvarrToMat to make the code portable between OpenCV 2 and 3:

Mat image_cv;
image_cv = cvarrToMat(static_cast<IplImage*>(image_yarp->getIplImage()));

[lornat75]

...and of course C++ style casts... I am getting old 👴