[FFmpeg-user] FFmpeg_Book.pdf 2.148

Michael Koch astroelectronic at t-online.de
Mon Jun 12 09:13:39 EEST 2023

Am 12.06.2023 um 00:00 schrieb Cecil Westerhof via ffmpeg-user:
> Michael Koch <astroelectronic at t-online.de> writes:
>> Am 11.06.2023 um 22:39 schrieb Cecil Westerhof via ffmpeg-user:
>>> In FFmpeg_Book.pdf chapter 2.148 is about video stabilisation.
>>> Part of the code is:
>>>       ffmpeg -i %IN% -vf vidstabdetect -y dummy.mov
>>>       del dummy.mov
>>>       ffmpeg -i %IN% -vf vidstabtransform -y %OUT%
>>> But line two removes what is generated in line one. So what are thos
>>> two lines doing?
>> It's described in the documentation for those two filters. The
>> vidstabdetect filter has two outputs. The stabilization data is written
>> to a file "transforms.trf", and the output of the filter chain is the
>> same as the input. Because this output isn't required, it can be
>> deleted.
>> The vidstabtransform filter in the third line does by default read the
>> file "transforms.trf".
> Ah, I understand. Thanks.
> Would it not be a good idea to also add a deshake example? Something
> like:
>      ffmpeg -i 00105.MTS -vf deshake=rx=64:ry=64:edge=3 -y deshake.mp4

There is a short example for deshake on the next page. But when I tested 
the deshake filter, I found that the two-pass solution with 
vidstabdetect and vidstabtransform gives much better results.
By the way, the vidstabdetect filter has a show=1 option which shows the 
motion vectors in the output video. In this case you need the output 
video from the first command line.


More information about the ffmpeg-user mailing list