Possible memory leak in VPI1.1 used in python

[erich.voko]

Hi,
I use VPI1.1 in python on my nano with live images from the HQ IMX477 camera.
When I use a simple example with resampling the image, the used RAM rises until 100% and from 80% onwards its getting very slow.
Is this a memory leak from VPI or I am doing something wrong? I’m new to VPI.
BR Erich

IMX477_VPI_test.py (4.2 KB)

[AastaLLL]

Hi,

Thanks for reporting this.
This is a known issue and our internal team is working on this:

Memory leak when using python VPI bindings

Hello. When I use a simple example like this: import os os.environ['OPENBLAS_CORETYPE'] = 'ARMV8' import cv2 import sys import vpi import numpy as np import time from argparse import ArgumentParser import gc backend = vpi.Backend.CUDA print('1') inVideo = cv2.VideoCapture(1) inVideo.set(cv2.CAP_PROP_FRAME_WIDTH, 480) inVideo.set(cv2.CAP_PROP_FRAME_HEIGHT, 640) inVideo.set(cv2.CAP_PROP_FPS, 30) if int(cv2.__version__.split('.')[0]) >= 3: extOutputVideo = '.mp4' fourcc = cv2.VideoWrit…

Will update more information with you once we got any feedback.
Thanks.

[erich.voko]

Great!
Hope for a fast fix because after appr. 5s I cannot use VPI anymore so its useless at the moment for live images.
BR Erich

[AastaLLL]

Hi,

While our internal is working on the solution, there is a WAR in the below comment:

Memory leak when using python VPI bindings

Hi, Thanks for your patience. Our internal team is working on this issue. Currently, you can use vpi.clear_cache() as a temporal solution to make sure the buffer is released. For example: while True: ret, cvFrame = inVideo.read() ... with vpi.Backend.CUDA: frame = vpi.asimage(cvFrame) ... vpi.clear_cache() Thanks.

Thanks.

[erich.voko]

Hi,
the vpi.clear_cache() helps!
Thanks for the fast workaround.

BR Erich