Documentation on how to make ffmpeg extract hq jpg images faster with hw de/encode

[jensdraht1999]

Documentation Not A Issue, So I Close Again.

The command should be changed to something like this on laptop with a igpu and dedicated gpu:

ffmpeg -hwaccel qsv -qsv_device 1 -c:v h264_qsv -i inputvideo.mp4 -threads 12 -vcodec mjpeg_qsv -qscale:v 2 output\%6doutput.jpeg

It was increasing from 30-32x to 37x extracting images on a 720p video. Need to lookup this info here later.

[jensdraht1999]

Here was the help I needed "https://www.reddit.com/r/ffmpeg/comments/100a8mf/error_thrown_when_trying_to_decode_with_qsv/"

[jensdraht1999]

Documentation 2: only for me..

• Cuda Image Extractor is slower than pure CPU frame extraction.
• Intel QSV has no scene change recognization, so it cannot extract frames from videos.
• We should not try to use faster lossless formats like QOI, because even when faster, they are not supported with FFMPEG/OPENCV/IMAGEMAGICK. So it is useless for now and probably will be for a long time.

[jensdraht1999]

Tested out, if QSV with CPU (two scripts running ffmpeg frame extract) is faster or not. It's not faster, qsv alone is much faster.