[MPlayer-users] Doubling frame rate

[Ivan Kalvachev]

Greg Trounson said:
> Gidday,
>
> I have a 30fps movie here that I would like to resample to 60fps, while
> keeping the motion real-time.  This will require twice as many frames as
> the original.
>
> I realise that I only have a limited number of frames to work with (half
> as many as I need), but I also know that MPEG has motion detection built
> in.  But that's about all I know.
>
> Is it possible to use mplayers motion estimation to generate
> intermediate frames for this kind of business, or am I being terribly naive?
>
> I tried:
> mencoder test30fps.avi -ovc lavc -lavcopts
> vcodec=mpeg4:vhq:vbitrate=4000 -oac copy -ofps 60 -o test60fps.avi
>
> but of course every second frame was reported as a duplicate frame.
>
> I know that algorithms exist to generate intermediate frames (hell, my
> 100Hz telly at home does it in realtime), but are they similar to what
> MPEG does?
>
> Any ideas?
>
> thanks,
> Greg
Well, that is very hard task.

First.
If your source material is from TV or camera capture, it is very probable
that it is been interlased. This mean that it contain 2 separate in time
fields. There is already reply with excact command which will help you
to split them and double the framerate. Be warned that it will hald the
horisontal resolution, so you may need to rescale or manually tune aspect.

Second.
There are many crapy 100Hz.
I've seen few. They all made motion blur. Very horrible.

Even if you/they use Motion Compensation, it is not enough because it is
usually block based (like in mpeg) and you will have defects on edges,
where one part of the edge goes one way, the other in opsosite way and
motion compensation should choose the best (but it cannot do that always)

MPEG-2 motion is even worse. It is not guaranteed to be an real motion
estamination. It just finds the "best match" block. I've seen an studio
encoded sample where motion vectors are like taken from /dev/random.

Final.
If you clip is short. Very short. Less than 10 frames ;)
You may try morphing. Of cource you may need to spend a lot of time
placing similarity points. But the result may be better.

Wish You Best
   Ivan Kalvachev
  iive