[MPlayer-users] Best inverse telecine filter for mixed content...

[Rich Felker]

On Thu, Jul 28, 2005 at 09:56:55PM -0700, RC wrote:
> This is spun-off of the thread "Crop before deinterlacing OK?"  The
> question is, which inverse telecine filter works best for TV captures,
> where you can't be sure if the content will be telecined or interlaced?
> 
> Here's one test, showing how many frames of
> interlaced material each inverse telecine filter outputs...
> 
> wc -l md5sums-* | sort -n              
>    798 md5sums-filmdint=dint_thres=256,pp=lb 
>    880 md5sums-ivtc,pp=lb 
>    881 md5sums-detc,pp=lb 
>    960 md5sums-pullup,pp=lb
>    998 md5sums-filmdint=io=1:1
>    998 md5sums-pp=lb
> 
> filmdint is noticably the worst for interlaced.  It's strictly
> enforcing the output of 24fps, even when the input is not telecined.
> Maybe that will look okay when encoding everything with -ofps
> 24000/1001 (I'll have to try that), but it looks quite bad for live

Agree, this will look bad.

> playback. Forcing it with io=1:1 is probably just making it output
> duplicate fields, which looks very bad when using those options with
> actual telecined material

It shouldn't duplicate anything for interlaced sequences. For
telecined content, yes, it will duplicate _frames_ (not fields).

> (may as well leave filmdint out).

Hell no. Then the frames won't be reconstructed at all and you'll have
horrible combing!

> detc/ivtc do a fairly good job, but I expected better.  Other than
> jumpiness, I never see defects from using detc/ivtc on interlaced
> materials.

Jumpiness is what you should expect to see..

> pullup seems to be doing a VERY good job, reducing the frame-rate by
> less than 4%.  Strict-breaks -1 didn't do any better.

Wow, cool. It was never intended to.. :)

> Okay, Rich, you can tell me why I'm completely wrong now...  ;-)

You're not, except about filmdint. Try
filmdint=io=1:1/dint_thres=256,pp=lb and see if you like it better.

Rich