Limit the ffmpeg to use as many cores as possible

Python
, ,
1

Previously I found a weird problem which I post on here to ask for some help:

Thanks for their instruction,however I still not figured it out how to set limitation on ffmpeg in cv2.videocapture（）function.
I use other way to handle this problem.
After building the opencv with gstreamer support this phenomenon might disappear.Also it has side-effect that the rtsp stream it read is not as fluent as ffmpeg although gstreamer is faster. still want to use the ffmpeg,can anyone give me some advice?

2

Have you tried setting the number of threads in the VideoCapture constructor as

cap = cv2.VideoCapture(addr, cv2.CAP_FFMPEG, [CAP_PROP_N_THREADS, nThreads ])
3

Thanks for your reply.You mean cap = cv2.VideoCapture(addr,cv2.CAP_FFMPEG, [cv2.CAP_PROP_N_THREADS, 2 ]) ? I tried this but cv2 seems dont’t have CAP_PROP_N_THREADS.
I also Tried like this cap = cv2.VideoCapture(addr,cv2.CAP_FFMPEG, [2, 2 ]) It failed ethier.

4

You need to build from a commit after this pr was merged

Upgrade to OpenCV 4.7.0.

1 Like
5

OK, you mean I should download from here GitHub - cudawarped/opencv at ffmpeg_rtsp_low_fps
and build opencv again.But how to check this pr’s opencv version?

6

Either install the latest pre-built wheel with pip

or build again checking out the 4.7.0 tag

after you have cloned the repository

7

OKOK.Thanks for your advice. By the way, if I want to limit the ffmepg to use less than 16 cores for example like 4 cores. The only way to do is to rebuild opencv4.7.0 right? Do you know how to change the default setting in opencv 4.7.0?

8

No. The option to change the number of threads is available at runtime, when you create VideoCapture as

cap = cv2.VideoCapture(addr, cv2.CAP_FFMPEG, [CAP_PROP_N_THREADS, 16])

You need to upgrade your version of OpenCV python bindings (build from source or pip install) because this feature was added in OpenCV 4.7.0 (is not available in opencv-python version : 4.6.0.66).

9

WOW! Thanks a lot. It seems like I am going to the wrong direction! Any way, THANKS!