[MPlayer-users] Best inverse telecine filter

[D Richard Felker III]

On Thu, Jan 08, 2004 at 08:39:12AM +0800, ephemeron at softhome.net wrote:
> [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> At least four different inverse telecine are listed in the
> mencoder man page: detc, ivtc, pullup and filmdint.  Which is the
> best?  I've ruled out detc (inferior to ivtc) and pullup
> (experimental).  So it's a tossup between -vf softpulldown,ivtc
> and -vf filmdint.
> 
> My limited tests tell me that filmdint is much faster than ivtc

The main reason is that softpulldown wastes a lot of time.

> but tends to produce video that "stutters" when the original is
> primarily hard telecined.  (The encoded video isn't just jerky.
> Some frames are encoded, it seems, in the wrong order, in a 1, 3,
> 2 sequence.)

Are you sure? I seriously doubt this. IMO it's more likely that it's
incorrectly deinterlacing rather than performing inverse telecine
(fimndint does both) which gives you flickering/choppy video.

> My important options for filmdint are "-vc mpeg12 -fps 29.970

-fps is never needed, and in fact will hurt a/v sync since the source
fps isn't  _exactly_ 29.97 but rather 30000/1001.

> -ofps 23.976 -vf filmdint=crop=W:H:X:Y" (I don't quite understand
> what the *thres=n options do) and for ivtc "-vc mpeg12 -ofps
> 23.976 -vf softpulldown,ivtc=1".  I'm looking for the best
> inverse telecine filter to stick into a script that will handle a
> variety of materials that are predominantly either soft or hard
> telecined or are mixed hard and soft telecined.  Right now I have
> to randomly examine parts of each video frame by frame (-fps 1
> -ss X:XX) to determine which filter to apply.

Sorry, it's hard to say in general... Some things are not meant to be
automated (without super-advanced AI networks...)

Rich