[FFmpeg-soc] [soc]: r1932 - libavfilter/vf_scale.c

[Bobby Bingham]

On Sun, 02 Mar 2008 18:11:10 +0100
Vitor Sessak <vitor1001 at gmail.com> wrote:

> Hi
> 
> Bobby Bingham wrote:
> > On Sun, 02 Mar 2008 15:00:34 +0100
> > Vitor Sessak <vitor1001 at gmail.com> wrote:
> >> Bobby Bingham wrote:
> >>> On Thu, 28 Feb 2008 00:18:59 +0100 (CET)
> >>> koorogi <subversion at mplayerhq.hu> wrote:
> >>>
> >>>> Author: koorogi
> >>>> Date: Thu Feb 28 00:18:58 2008
> >>>> New Revision: 1932
> >>>>
> >>>> Log:
> >>>> Slice support for vf_scale.
> >>>>
> >>>>
> >>>> Modified:
> >>>>    libavfilter/vf_scale.c
> >>>>
> >> [...]
> >>
> >>> After some more testing, it seems this doesn't work right with
> >>> --enable-swscaler. I'll take a closer look at it later tonight.
> >> I was wondering one thing: imagine doing a 200x200 -> 400x400
> >> linear interpolation resizing.
> >>
> >> The filter gets as input a (for ex.) 4x100 slice. It can then 
> >> interpolate the line numbers (1,2,3,4,5,6,7) of the output. For the
> >> next slice, it would be able to output the lines
> >> (9,10,11,12,13,14,15). But to evaluate the output line 8, it'll
> >> need both the input line 4 and 5. So to do slice-based resizing,
> >> it would be necessary a complicated, resize algorithm dependent,
> >> caching. Also it would require some kind of memcpy.
> >>
> >> So my opinion is that implementing slices in resizing is hard and
> >> maybe not worth the effort. In the other hand, colospace conversion
> >> should be pretty simple. Or am I missing something?
> >>
> > 
> > I believe that this caching is already implemented by libswscale.
> 
> Does imgresample.c also support this?

It doesn't appear to. Hmmm...

Maybe it's a good time to implement some flags to let filters tell the
system that they require full-frame slices, and make vf_scale use that
in the imgresample.c case.

-- 
Bobby Bingham
Never trust atoms.  Or anything made of atoms.
このメールは再利用されたバイトでできている。