将FFMPEG的GPU版本编译为OpenCV

Question

When I try to use GPU FFMPEG on opencv by using a command like:

VideoCapture cap("rtsp link");

for(;;)
{
mat frame;
cap>>frame;
}

I don't see any output on nvidia-smi despite compiling the GPU version of FFMPEG.

I am trying to get FFMPEG to use GPU to improve the speed of my program. I got FFMPEG from cloning this and followed the steps here to get it working.

I have validated the results using nvidia-smi dmon -i 0 and have seen that dec% increases when I run the command ffmpeg -c:v h264_cuvid -i test_video.mp4 test_output.mkv .

I then compiled Opencv to use the GPU enabled FFMPEG, I check this by running getBuildInformation from opencv:

Video I/O: 
DC1394 1.x: NO 
DC1394 2.x: YES (ver 2.2.4) 
FFMPEG: YES 
codec: YES (ver 58.20.103) 
format: YES (ver 58.17.100) 
util: YES (ver 56.18.102) 
swscale: YES (ver 5.2.100) 
resample: NO 
gentoo-style: YES 
GStreamer: NO 
OpenNI: NO 
OpenNI PrimeSensor Modules: NO 
OpenNI2: NO 
PvAPI: NO 
GigEVisionSDK: NO 
UniCap: NO 
UniCap ucil: NO 
V4L/V4L2: NO/YES 
XIMEA: NO 
Xine: NO 
gPhoto2: NO

I cross-verify that I'm using the same ffmpeg (by checking the version numbers):

ffmpeg version N-91273-g67747c8 Copyright (c) 2000-2018 the FFmpeg developers
  built with gcc 5.4.0 (Ubuntu 5.4.0-6ubuntu1~16.04.9) 20160609
  configuration: --enable-cuda --enable-cuvid --enable-nvenc --enable-nonfree --enable-libnpp --extra-cflags=-I/usr/local/cuda/include --extra-ldflags=-L/usr/local/cuda/lib64
  libavutil      56. 18.102 / 56. 18.102
  libavcodec     58. 20.101 / 58. 20.101
  libavformat    58. 17.100 / 58. 17.100
  libavdevice    58.  4.101 / 58.  4.101
  libavfilter     7. 25.100 /  7. 25.100
  libswscale      5.  2.100 /  5.  2.100
  libswresample   3.  2.100 /  3.  2.100

Please advise me on how I should compile OpenCV to use the GPU version of FFMPEG. Thank you!

Answer 1

Not strictly FFmpeg related, but as an advise for you to be able to use hugely performant GPU capabilities through OpenCV API.

Assuming that you use OpenCV 3.xx - you should use special OpenCV GPU Api for video decoding. Take a look on this example from official documentation:

• https://github.com/opencv/opencv/blob/master/samples/gpu/video_reader.cpp

Also, foreseeing the problems with compilation and running the code with modern OpenCV 3.4.1 and >=CUDA 9.0, read through my comment on OpenCV's issue tracker to overcome those:

• https://github.com/opencv/opencv/issues/10201#issuecomment-402453361

And also yes - it will give you huge speed-up on video decoding/reading. Absolutely worth trying.