Object tracking: GOTURN - failed to allocate 90 GB cv::OutOfMemoryError

asked 2019-12-09 05:06:55 -0600

Vytau
46 ●4

updated 2019-12-09 07:57:41 -0600

Akhil Patel
239 ●6 ●33 ●40

I am trying to implement the GOTURN object tracking algorithm as described in this link. The error occurs as the tracker gets initiallized with the first image is passed over:

bbox = cv2.selectROI(frame, False)
ok, frame = video.read()
ok = tracker.init(frame, bbox)

cv2.error: OpenCV(4.1.2-dev) \opencv\modules\core\src\alloc.cpp:73: error: (-4:Insufficient memory)
Failed to allocate 95588562276 bytes in function 'cv::OutOfMemoryError'

Well, apparently, the algorithm tries to allocate almost 90 GB of memory, which is somehow crazy. I use the caffe model and the prototxt files from here.

Does anyone have any idea?

edit retag flag offensive close merge delete

Comments

I have no idea, but funny things tend to happen with bad data, so I'd check the sanity of the frame and bbox.

mvuori ( 2019-12-09 09:10:09 -0600 )edit

i tried on colab , 25gb ram wasn't enough ;(

edit: it seems to work with 3.4.8, so it's a regression.

i think, you should raise an issue here

berak ( 2019-12-11 04:32:05 -0600 )edit

we can also reproduce it with a simple:

net = cv2.dnn.readNet("goturn.caffemodel", "goturn.prototxt")

berak ( 2019-12-11 05:38:16 -0600 )edit

@berak. The code will work under 3.x.x. Unfortunately doesn't work OpenCV 4.x.x. Fortunaltey, I managed to get it working for merely OpenCV 4.

Could not find GOTURN model in current directory.
Please ensure goturn.caffemodel and goturn.prototxt are in the current directory.
Select a ROI and then press SPACE or ENTER button!
Cancel the selection process by pressing c button!

Snippet:

 if not(os.path.isfile('goturn.caffemodel') and os.path.isfile('goturn.prototxt')):
         errorMsg = '''
         Could not find GOTURN model in current directory.
         Please ensure goturn.caffemodel and goturn.prototxt are in the current directory
         '''
         print(errorMsg)
         sys.exit()

supra56 ( 2019-12-11 07:41:18 -0600 )edit

@supra56, imho, the problem is a regression in the dnn, not the tracker code (or python specific)

berak ( 2019-12-11 07:46:24 -0600 )edit

@berak Thank you. I guess I am gonna have to open the issue...

Vytau ( 2019-12-11 08:58:46 -0600 )edit

any updates on this ? shall i move to a less recent version of opencv?

aya sal ( 2020-01-16 15:59:19 -0600 )edit

@aya sal -- unfortunately, no updates or fix so far.

berak ( 2020-01-17 01:46:25 -0600 )edit

You can look at this issue. It can help with most of the cases, but still, it is not working properly if the object is going out of the picture.

ieliz ( 2020-07-08 06:42:25 -0600 )edit

add a comment

Object tracking: GOTURN - failed to allocate 90 GB cv::OutOfMemoryError

Comments

1 answer

Links

Question Tools

Stats

Related questions

Object tracking: GOTURN - failed to allocate 90 GB cv::OutOfMemoryError edit

Comments

1 answer

Links

Question Tools

Stats

Related questions

Object tracking: GOTURN - failed to allocate 90 GB cv::OutOfMemoryError