[FFmpeg-devel] [PATCH] simplify crop/pad handling in ffmpeg.c
Hervé
herve.flores
Wed Apr 16 14:06:29 CEST 2008
Le 16 avr. 08 ? 13:38, Luca Abeni a ?crit :
> Hi,
>
> Diego Biurrun wrote:
>> This is a patch from Herv? Flores to simplify crop/pad handling in
>> ffmpeg.c. For your consideration.
> [...]
>
>> Index: ffmpeg.c
>> ===================================================================
>> --- ffmpeg.c (revision 12847)
>> +++ ffmpeg.c (working copy)
>> @@ -2262,10 +2262,6 @@
>> fprintf(stderr, "Incorrect top crop size\n");
>> av_exit(1);
>> }
>> - if ((frame_topBand % 2) != 0) {
>> - fprintf(stderr, "Top crop size must be a multiple of 2\n");
>> - av_exit(1);
>> - }
>
> Sorry; maybe I am missing something, but I see this check removed
> from here
> and not re-introduced in other places... Do we support cropping and
> padding
> with non-multiple-of-2 quantities?
the idea of this patch (not from me but from a guy from pixar, sent in
the list a long time ago) was to permit an odd value for pad and crop.
If your file has an odd height, ffmpeg does not handle it
with this patch you can pad (or crop) only one pixel to have a final
even value
usually, each parameter check for even value an exit if value is odd
with this patch, the only check is done with the resulting value :
- if height - crop + pad = even, so no problem
- if height - crop + pad = odd, exit
PS: I don't remember exactly what was the pb and how to reproduce it,
but if I well remeber, this patch have a pb with chroma and interlaced
streams (bad resample, no "swap"/reinterlacing of the chroma channel)
eg: file has an even height, if you -padbottom 1 and -padtop 1 => no
good chroma resample
another bug with this patch: file has an even height, if you -
cropbottom 1 and -padbottom 1 => no change (no pad, no crop)
I think the easiest way to correct this patch will be: final check in
the patch shouldn't permit odd pad/crop if the file have a native even
value => exit
(PS: I'm not a C coder, I just manage to read it a little, that's all)
bye
Herv?
More information about the ffmpeg-devel
mailing list