QImage内存泄漏

Question

I've written a OpenCV application which basically grabs frames from a camera, does some image processing and displays the image in two edited variants. First, I've used cv2.imshow() to display the images, but while OpenCV (Build without Qt support) isn't able to provide modern GUI elements, I decided to use PySide for my GUI.

But since this I get this error after processing about 830-850 frames (no matter, what timer rate I use, or how much image processing I do):

QImage: out of memory, returning null image

for both of my image views in the GUI, and then in each loop this one:

OpenCV Error: Unspecified error (The numpy array of typenum=2, ndims=3 can not be created) in NumpyAllocator::allocate, file ..\..\..\opencv-3.1.0\modules\python\src2\cv2.cpp, line 184
OpenCV Error: Insufficient memory (Failed to allocate 921600 bytes) in cv::OutOfMemoryError, file ..\..\..\opencv-3.1.0\modules\core\src\alloc.cpp, line 52
Traceback (most recent call last):
  File "C:/myfile.py", line 140, in process_frame
    img = QtGui.QImage(cv2.cvtColor(thresh_img, cv2.COLOR_RGB2BGR), self.width, self.height,
cv2.error: ..\..\..\opencv-3.1.0\modules\core\src\alloc.cpp:52: error: (-4) Failed to allocate 921600 bytes in function cv::OutOfMemoryError

Here's a part of my code (without image processing, but it also produces the error):

import cv2
import sys
from PySide import QtGui, QtCore
from threading import Thread


class MainWindow(QtGui.QMainWindow):
    def __init__(self, cam=0, parent=None):
        super(MainWindow, self).__init__(parent)

        self.camera = Camera(cam).start()
        self.title = "Cam %s" % cam
        self.counter = 0

        widget = QtGui.QWidget()
        self.layout = QtGui.QBoxLayout(QtGui.QBoxLayout.LeftToRight)

        self.video_frame = QtGui.QLabel()
        self.thresh_frame = QtGui.QLabel()

        self.layout.addWidget(self.video_frame)
        self.layout.addWidget(self.thresh_frame)
        self.layout.addStretch()

        self.setCentralWidget(widget)
        widget.setLayout(self.layout)

        self.setMinimumSize(640, 480)
        self._timer = QtCore.QTimer(self)
        self._timer.timeout.connect(self.process_frame)
        self._timer.start(20)

    def process_frame(self):
        self.counter += 1
        print(self.counter)
        self.frame = self.camera.read()
        self.height, self.width = self.frame.shape[:2]

        thresh_img = cv2.threshold(cv2.cvtColor(self.frame, cv2.COLOR_RGB2GRAY), 0, 255,
                                   cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
        thresh_img = cv2.erode(thresh_img, None, iterations=2)
        thresh_img = cv2.dilate(thresh_img, None, iterations=2)
        thresh_img = cv2.cvtColor(thresh_img, cv2.COLOR_GRAY2RGB)

        img = QtGui.QImage(cv2.cvtColor(self.frame, cv2.COLOR_RGB2BGR), self.width, self.height,
                           QtGui.QImage.Format_RGB888)
        img = QtGui.QPixmap.fromImage(img)
        self.video_frame.setPixmap(img)

        img = QtGui.QImage(cv2.cvtColor(thresh_img, cv2.COLOR_RGB2BGR), self.width, self.height,
                           QtGui.QImage.Format_RGB888)
        img = QtGui.QPixmap.fromImage(img)
        self.thresh_frame.setPixmap(img)

    def closeEvent(self, event):
        self.camera.stop()
        event.accept()


class Camera:
    def __init__(self, src=0):
        self.stream = cv2.VideoCapture(src)
        (self.grabbed, self.frame) = self.stream.read()

        self.stopped = False

    def start(self):
        Thread(target=self.update, args=()).start()
        return self

    def update(self):
        while True:
            if self.stopped:
                return
            (self.grabbed, self.frame) = self.stream.read()

    def read(self):
        return self.frame

    def stop(self):
        self.stopped = True

if __name__ == "__main__":
    app = QtGui.QApplication(sys.argv)
    window = MainWindow(0)
    window.show()
    sys.exit(app.exec_())

In the Windows task manager I can see the RAM usage of my program:

At the point of crash, the app uses about 1.5 GB of RAM. I've tried using the gc module and gc.collect() after del img , no success.

What else can I do?

EDIT:

The threaded Camera class doesn't matter here, the error does also appear without it.

Answer 1

It seems to be a PySide specific bug, using PyQt will fix it. It's not even OpenCV related. It doesn't look like there will be a solution for using PySide right now...

Answer 2

Apparently this bug is due to garbage collection issues when using python 3.x

A simple workaround using ctypes has been provided here https://bugreports.qt.io/browse/PYSIDE-140 and https://github.com/matplotlib/matplotlib/issues/4283#issuecomment-92773487

On the last link see the post "mfitzp commented on Apr 24, 2015". This worked for me!

Answer 3

In my case, I was using QStackedWidget for switching between different views and QTimer to trigger the view switch (on certain conditions).

I was using functools.partial for passing arguments from one view to another, among which I did pass instances, such as PIL.Image , QImage , ImageQt and QPixmap .

That's where it went wrong. When such resources are passed from one view to another, they are not being cleaned well by the garbage collector.

What worked for me.

1. Declare all variables that you plan to pass from one view to another as class properties (in the constructor of your QWidget ). Store all image resources (such as PIL.Image , QImage , ImageQt and QPixmap ) in those variables/properties.

2. Don't pass any arguments (containing image resources) from one view to another.

---

Additionally if you're using PyQT with OpenCV, note that OpenCV 2.9 contains a bug, which causes memory leak. Make sure to upgrade to latest version (either latest 2.x or 3.x) before you spend hours on finding out where the memory leaks come from.

Answer 4

Look at my comment at PYSIDE-140: https://bugreports.qt.io/browse/PYSIDE-140

Avoiding mixing python objects with Qt objects. Try example below and avoid playing with refcounts fixing.

stream = cv2.VideoCapture(0)
grabbed, frame = stream.read()

height, width = frame.shape[:2]
data = Qt.QByteArray(cv2.cvtColor(frame, cv2.COLOR_RGB2BGR).tostring())
qimg = Qt.QImage(data, width, height, Qt.QImage.Format_RGB888)

data.clear()
del data