[FFmpeg-user] FFMpeg problem using fieldmatch and decimate filters to IVTC

[L. Lee]

On 5/15/14, 9:06 AM, "Carl Eugen Hoyos" <cehoyos at ag.or.at> wrote:

>L. Lee <llee040 <at> sbcglobal.net> writes:
>
>> ffmpeg -i 'long.sample.vob'
>
>> -vf "fieldmatch=order=tff, fps=fps=24000/1001"
>
>This is definitely not supposed to work (and if
>you found documentation somewhere suggesting
>this, please tell us!)
>
>If you have a video file that cannot be inverse
>telecined with "-vf pullup -r 24000/1001" please
>provide it!
>

One of the benefits of doing this as a hobby rather than as a means of
seeking to provide a source of income, I guess, is that I¹m interested in
trying whatever combinations I can find that will produce an output file
from an input file. However, in this case, I simply combined ffmpeg
documented examples of the use of the fps option with the filtergraph for
detelecine, and with Steve¹s example, the output is much closer to what is
desirable than what I get by applying ³-vf pullup -r 24000/1001².

But I certainly don¹t intend to get into a debate about this. I don¹t
claim to be an authority on this sort of transcoding process. Steve made
his examples publicly available, and anyone can download and test my
suggested filter commands for inverse telecine with the latest build of
ffmpeg on any system and compare results to those achieved through the use
of a conventional method. If you see an output video file created using
one of my suggestions that looks smoother than what you get when you use
³-vf pullup -r 24000/1001² then use mplayer or something else that will
let you step through the output one frame at a time as well as analyze the
statistical data. That will hopefully provide information about what is
happening and why, and whether it is acceptable as well as whether methods
of improving the strategy may be needed.

I still believe that the information and evidence I¹ve provided supports
the assertion that ffmpeg has not yet attained the level of success for
applying inverse telecine to video that is possible through the use of
MEncoder. Use my previously posted MEncoder command line suggestions on
the contributed source material to compare with the results of ffmpeg¹s
³-vf pullup -r 24000/1001² or any of the others that I¹ve already
submitted and judge for yourself. If you have trouble obtaining Steve¹s
sample passage, I¹m sure he will be willing to help find a way to
accommodate you.

And having said all that, I have one more filtergraph suggestion that
seems to work about as well than any of those I previously suggested
(although it doesn¹t IMO approach the level of success achieved by using
the MEncoder conversion that I submitted), but this one definitely leaves
interlaced frames. The interesting thing is that it works at all and
achieves a different result from others. I think I may be able to find a
way to remove the remnants of interlacing eventually.

ffmpeg -i long.sample.vob  -vf "fps=60001/1001, fieldmatch=order=tff,
fps=fps=24000/1001" -pix_fmt yuv420p -c:v libx264 -preset slower -tune
film -crf 18 -maxrate 2100k -bufsize 1835k long.sample.m4v


Thanks.

Laine Lee