[FFmpeg-devel] [PATCH] lavf/apngdec: use AVStream.time_base instead of r_frame_rate

[Benoit Fouet]

Hi,

On November 25, 2014 10:15:31 PM GMT+01:00, Michael Niedermayer <michaelni at gmx.at> wrote:
>On Tue, Nov 25, 2014 at 02:56:07PM -0300, James Almer wrote:
>> Should fix framedrops on some apng files
>> 
>> Signed-off-by: James Almer <jamrial at gmail.com>
>> ---
>> This is still not optimal because the value of time_base will be
>updated on every frame, 
>> and in some cases delay_num and delay_den varies between frames.
>> 
>> Better fix welcome.
>> 
>>  libavformat/apngdec.c | 4 ++--
>>  1 file changed, 2 insertions(+), 2 deletions(-)
>> 
>> diff --git a/libavformat/apngdec.c b/libavformat/apngdec.c
>> index d766a87..47d3753 100644
>> --- a/libavformat/apngdec.c
>> +++ b/libavformat/apngdec.c
>> @@ -265,8 +265,8 @@ static int decode_fctl_chunk(AVFormatContext *s,
>APNGDemuxContext *ctx, AVPacket
>>          delay_num = 1;
>>          delay_den = ctx->default_fps;
>>      }
>> -    s->streams[0]->r_frame_rate.num = delay_den;
>> -    s->streams[0]->r_frame_rate.den = delay_num;
>> +    s->streams[0]->time_base.num = delay_num;
>> +    s->streams[0]->time_base.den = delay_den;
>>      pkt->duration = 1;
>
>this is wrong, the timebase and r_frame_rate are constant
>once they have been set
>

OK, good to know.

>i suggest to use set_pts_info to set a reasonable precisse
>timebase something like one millionth or billionth or similar in
>base 2. and then set pts based on these delays and the previous
>timestamp

James, do you want to have a look at this approach or do you want me to do it?

>That is unless apng has proper timestamps, in which case they
>should be used
>

It doesn't. Each frame has its duration, which is represented by a fraction of a second, and it can very well change between images.

-- 
Ben