On Fri, 20 Apr 2018, Michael Niedermayer wrote:
On Thu, Apr 19, 2018 at 09:32:19PM +0200, Marton Balint wrote:A pixel format which has a palette also has alpha, without this patch the format negotiation might have preferred formats without alpha even if the source had alpha. Signed-off-by: Marton Balint <c...@passwd.hu> --- libavfilter/avfiltergraph.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/libavfilter/avfiltergraph.c b/libavfilter/avfiltergraph.c index 4cc6892404..e18f733e23 100644 --- a/libavfilter/avfiltergraph.c +++ b/libavfilter/avfiltergraph.c @@ -679,7 +679,7 @@ static int pick_format(AVFilterLink *link, AVFilterLink *ref) if (link->type == AVMEDIA_TYPE_VIDEO) { if(ref && ref->type == AVMEDIA_TYPE_VIDEO){ - int has_alpha= av_pix_fmt_desc_get(ref->format)->nb_components % 2 == 0; + int has_alpha= av_pix_fmt_desc_get(ref->format)->nb_components % 2 == 0 || (av_pix_fmt_desc_get(ref->format)->flags & AV_PIX_FMT_FLAG_PAL);This causes various output files to grow in size with a unused alpha plane a random example: (there are likels better examples) ./ffmpeg -y -i ~/tickets/1116/1023.bmp -vf negate fileX.bmp Iam not sure unconditionally treating all palettes as if they have non fully opaque entries is a good idea.
Obviously not, but it is already treated this way in most places. Having a bigger image with alpha is better than losing alpha. And the user can always force losing alpha with a filter, and sometimes he has to do that anyway because for example fre0r filters have no way of signalling if they use alpha or not...
Doesnt this patchset replace a problem by another problem ? It may be better to not rush this and find a complete solution a 2nd pixfmt and or scaning the 256 entries for their alpha values is maybe the direction we could go
I don't think scaning can work for anything other than still images, as the palette can change per-frame AFAIK. Introducing a new pixel format seems the better approach, however keeping in mind compatibility and existing code, I'd rather make the newly introduced one the one without alpha, and move things gradually to it. Still seems like a fair amount of work...
Regards, Marton _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
