[FFmpeg-devel] [PATCH] change frame_aspect_ratio to AVRational

[Baptiste Coudurier]

On 6/11/2009 3:34 PM, Michael Niedermayer wrote:
> On Thu, Jun 11, 2009 at 03:06:54PM -0700, Baptiste Coudurier wrote:
>> On 6/11/2009 1:14 PM, Michael Niedermayer wrote:
>>> On Thu, Jun 11, 2009 at 11:36:31AM -0700, Baptiste Coudurier wrote:
>>>> On 6/11/2009 4:27 AM, Michael Niedermayer wrote:
>>>>> On Wed, Jun 10, 2009 at 09:23:40PM -0700, Baptiste Coudurier wrote:
>>>>>> Hi
>>>>>>
>>>>>> $subject, mainly to avoid using double which behaves not correctly
>>>>>> when using 16:9 commandline.
>>>>> [...]
>>>>>> @@ -3012,7 +3012,8 @@
>>>>>>          st->stream_copy = 1;
>>>>>>          video_enc->codec_type = CODEC_TYPE_VIDEO;
>>>>>>          video_enc->sample_aspect_ratio =
>>>>>> -        st->sample_aspect_ratio = av_d2q(frame_aspect_ratio*frame_height/frame_width, 255);
>>>>>> +        st->sample_aspect_ratio =
>>>>>> +            av_mul_q(frame_aspect_ratio, (AVRational){ frame_height, frame_width });
>>>>>>      } else {
>>>>>>          const char *p;
>>>>>>          int i;
>>>>>> @@ -3039,7 +3040,8 @@
>>>>>>  
>>>>>>          video_enc->width = frame_width + frame_padright + frame_padleft;
>>>>>>          video_enc->height = frame_height + frame_padtop + frame_padbottom;
>>>>>> -        video_enc->sample_aspect_ratio = av_d2q(frame_aspect_ratio*video_enc->height/video_enc->width, 255);
>>>>>> +        video_enc->sample_aspect_ratio =
>>>>>> +            av_mul_q(frame_aspect_ratio, (AVRational){ frame_height, frame_width });
>>>>>>          video_enc->pix_fmt = frame_pix_fmt;
>>>>>>          st->sample_aspect_ratio = video_enc->sample_aspect_ratio;
>>>>>>  
>>>>> previously these where limited to 255/255 now they arent anymore, i think
>>>>> some codecs dont support arbitrary fractions
>>>> Ok, quick survey:
>>>> theora encoder supports rational.
>>>> h264 supports rational on 16bits.
>>>> libxvid indeed checks for 255/255.
>>>> mpeg12 converts back to float and match supported ones.
>>>> mjpeg is rational on 16bits.
>>>> h263/mpeg4 sar is rational on 8bits.
>>>>
>>>> Ok to add av_reduce to codecs needing it ?
>>> encoders should not second guess the user
>>> i mean if they accept a 300/301 then they should either store that or
>>> if they cannot then fail
>>>
>>> it seems to me the current code is fine, is your patch fixing something
>>> besides changing float to AVRational ?
>>> if not we maybe could just keep it as it is
>> Yes it keeps current par when converting when par > 255.
>> resolution 720x512 DAR 16:9 PAR 512:405.
>>
>> mov can store this perfectly, codec par must be equal to stream par
>> and current code alters both.
>>
>> Stream #0.0: Video: mpeg2video, yuv422p, 720x512 [PAR 67:53 DAR
>> 3015:1696], q=2-31, 30000 kb/s, 90k tbn, 29.97 tbc
>>
>> Even with stream copy, mov can store rational on 32bits.
>> and H.264 supports this exact PAR corresponding to DAR.
>>
>> Can we please apply the patch ?
> 
> hmm, iam not entirely happy with that patch
> at least av_reduce() has to be added where its needed
> and that needs to take care of cases where a non 0 value becomes
> 0 through av_reduce()
> maybe we could add a max_sar_component to AVCodec and use that to
> reduce in ffmpeg.c ?
> It feels correcter, i mean
> "lets set 1000/513 and see what comes out, never knowing what really
>  is stored ..."
> vs.
> "Ahh codec supports up to 255/255, lets reduce and store that"
> 
> PS: We already have seperate sar in AVStream and AVCodecContext so it
> would not affect muxers

Well, if there is a way to have a max rational value for
"sample_aspect_ratio" in AVOption (255/255), I can implement per codec
defaults and min/max using AVOptions * in AVCodec, what do you think ?

-- 
Baptiste COUDURIER                              GnuPG Key Id: 0x5C1ABAAA
Key fingerprint                 8D77134D20CC9220201FC5DB0AC9325C5C1ABAAA
FFmpeg maintainer                                  http://www.ffmpeg.org