[FFmpeg-user] Extracting frames from 59.94 FPS video - only every other frame is different

Rob Hallam ffmpeg at roberthallam.com
Fri Dec 1 22:03:46 EET 2017

On 1 December 2017 at 17:12, Will Price <will.price at bristol.ac.uk> wrote:

> I've used `compare` from ImageMagick to actually determine whether the frames are different, and they are, the diff looks like a frame itself but with high brightness. (see frames 1, 2, and the diff at this link: https://drive.google.com/open?id=1Tn3dbZ6Ss7OdhHh6OiJT4qsPXSqCVVgF). If you visually compare frames 1 and 2 then to the human eye they look impercetibly different, yet the diff shows that there is a difference. When I look at the frames in a photo viewer and iterate through them rapidly the apparent motion is much more jerky than in the source video.

I may be barking up the wrong tree, but do you expect to see a
perceptible difference between frames 1 and 2?

I haven't played much with IM's `compare`, but it looks like you are
in an office with possibly fluorescent lights - could the difference
in brightness be different parts of the 50Hz flicker?

> The motivation for needing smooth non-duplicated frames is because they are to be used as input to an optical flow algorithm which then produces inaccurate optical flow estimations due to the similarity between pairs of frames starting with an odd frame number.

Do you need 1920x1080x60 resolution? Would removing the 'similar'
images work? Would increasing the frequency to 120FPS do the trick?Or
higher? Seems like the Hero 5 has access to HEVC
(https://gopro.com/help/articles/block/hevc), is the frame output
issue there as well?

Someone more knowledgeable than I on h264 can perhaps weigh in on how
GoPro record and encode their video. Given the constraints of popular
consumer-level (relatively cheap) hardware aimed at sports / outdoor
activities etc, I wouldn't be surprised if they were doing something
with capture/h264 encoding that gives the behaviour you describe
and/or makes it unsuitable for your purposes.

For confirmation, you could compare the output ffmpeg supplies you
with GoPro's own 'Studio' software:



> ffmpeg version 3.4

You will possibly be told that 'latest git head' is what is officially
supported on the ML!

More information about the ffmpeg-user mailing list