Julia Opencv memory leak

General Usage
1

Hello. Good afternoon.
Im using OpenCV.jl and Im facing a memory leak using video capture in this example. May someone help?

julia 1.11.3 Win x64

image

using OpenCV

    cap = OpenCV.VideoCapture(0)
    if !OpenCV.isOpened(cap)
        error("camera not found")
    end

    OpenCV.namedWindow("Webcam")

    try
        while true
            ret, frame = OpenCV.read(cap)
            if !ret
                println("Não foi possível capturar o frame")
                break
            end
            OpenCV.imshow("Webcam", frame)
            if OpenCV.waitKey(33) == Int('q')
                break
            end

        end
    finally
        OpenCV.release(cap)
        OpenCV.destroyAllWindows()
    end
2

What do you observe that makes you think you are facing a memory leak? You run the code, see the window with an image from your webcam, and overtime the memory usage grows? Does the amount of used memory increase until you get an out-of-memory error?

I suspect that there is no memory leak and that the GC just hasn’t collected the frames that are no longer referenced. To test that hypothesis – if you put GC.gc() right at the end of the while loop, does that fix it?

1 Like
3

Hello, good morning. Yes. Using gc it is collected. What do you mean, is that julia will let memory increase more than 5Gb? I will do the tests e let it run for thirthy minutes

4

yep. It frozen my machine at 96% total of memory. Its a leak. Using GC. gc() really collects the memory, but, julia dont know how to collect without calling gc explicitly.

5

What do you mean, is that julia will let memory increase more than 5Gb?

My knowledge about GC is very limited. The documentation says that

The GC will do full collections when the heap size reaches 80% of the maximum allowed size.

I am not sure what the maximum allowed size is, but it is certainly no more than the amount of physical memory, so I am not sure why the GC does not run in your case.

What if you you pass --heap-size-hint=1GB (or another value) to julia when you start your script? That should force garbage collection earlier.

1 Like
6

I will try this test

7

Dont stop

image
image1597×617 54.5 KB

According this, its an old issue

2 Likes
8

Oh wow, that’s unfortunate. I see that opencv provides Mat::release but it is not available in the wrapper. I guess you are stuck with calling GC.gc() and the occasional delays that come with it.

1 Like
9

Yep. If I call it every time, will make the program perform very bad. Buts ok. I can do the job only using static images for a while.